2. Prosodic domains in French

In QF, as in other varieties of French, prominence is obligatorily assigned only at the right edge of a PPh. As a result, in a PPh that has more than one lexical word, each of which is parsed as a PWd, only the rightmost PWd is assigned prominence (see, e.g., Dell Reference Dell, Dell and Vergnaud1984 and Jun & Fougeron Reference Jun, Fougeron and Botinis2000 for European French and Thibault & Ouellet Reference Thibault, Ouellet, Bunnell and Idsardi1996 for QF). The example in (1) shows that DPs with prenominal adjectives are parsed as one PPh in French (Jun & Fougeron Reference Jun, Fougeron and Botinis2000): thus, only the final syllable of maison is prominent, indicated in bold. Final prominence is acoustically manifested through longer duration and higher pitch relative to non-prominent syllables (e.g., Jun & Fougeron Reference Jun, Fougeron and Botinis2000 for European French and Lamontagne & Goad Reference Lamontagne and Goad2022 for QF).

In most varieties of French, including QF, an optional H tone can be placed on the first syllable of the first lexical word (or PWd) of the PPh (Thibault & Ouellet Reference Thibault, Ouellet, Bunnell and Idsardi1996 for QF and Jun & Fougeron Reference Jun, Fougeron and Botinis2000 for European French). This means that in PPhs with a single long word, the optional left-edge H tone can fall on the same PWd that is assigned the obligatory final prominence, the phrasal high tone indicated by H* in (2a). In contrast, in PPhs with two PWds, each PWd may be assigned a H tone, as shown in (2b).

In the literature on French, there is some disagreement on the inventory of prosodic domains employed in the language. Jun & Fougeron (Reference Jun, Fougeron, Elenius and Branderud1995, Reference Jun, Fougeron and Botinis2000, Reference Jun and Fougeron2002) exclude both the PWd and the foot from the hierarchical structure of domains they posit: syllables are organised into function words or lexical words which are, in turn, organised into Accentual Phrases (equivalent to PPh, as mentioned above). We will argue that HVD is sensitive to iteratively built feet which are combined into the PWd, as per the prosodic hierarchy; the domain in which HVD is computed is thus the PWd. In view of this, we briefly discuss the status of each of these constituents in French from earlier literature.

Jun & Fougeron’s (Reference Jun, Fougeron and Botinis2000) research focuses on prominence-related phenomena. In works which examine constraints on syllable and segmental well-formedness, empirical support has been provided for the PWd. For example, glide formation is observed PWd-internally in French, as shown in (3a), but not across PWd boundaries, as shown in (3b) (Hannahs Reference Hannahs1995; Lyche & Girard Reference Lyche and Girard1995). That the relevant domain is the PWd rather than the lexical word can be seen from the behaviour of derivationally complex forms: suffixes that fall inside the PWd of their host trigger glide formation, as in (3c); while prefixes that form a PWd as a sister to the PWd of their host do not, as in (3d) (following Guzzo Reference Guzzo2018).

Turning to the foot, evidence for this constituent in French is less clear. Aside from the absence of lexical stress, French does not have a minimal word requirement. In languages with feet, the minimal lexical word usually corresponds to a binary (bimoraic or disyllabic) foot (McCarthy & Prince Reference McCarthy and Prince1990, Reference McCarthy and Prince1995). Although French has a vowel length distinction, all oral vowels are realised as short word-finally (e.g., Walker Reference Walker1984; Montreuil Reference Montreuil and Lyche1995; Côté Reference Côté, Gess, Lyche and Meisenburg2012) and, as a result, numerous CV lexical words are attested (e.g., Scullen Reference Scullen1997); see (4).

In contrast to this view, several researchers have proposed that French does have foot structure. Selkirk (Reference Selkirk1978) takes this position to account for patterns in the distribution of schwa. She proposes that every syllable with a full vowel corresponds to its own foot, while a syllable containing a schwa, which is prosodically deficient, attaches to the syllable to its left, forming a trochee. A structure such as illustrates this parsing. This proposal in Selkirk (Reference Selkirk1978) has been further developed in subsequent analyses, such as Montreuil (Reference Montreuil and Lyche1995, Reference Montreuil, Agwuele and Park2002) and Bosworth (Reference Bosworth2017). A trochaic approach has also been proposed to account for the stress patterns observed in Early Old French: while word-final consonants in stressed syllables are taken to be onsets of empty-headed syllables, , word-final syllables with a schwa are analysed as being in foot-dependent position, (Rainsford Reference Rainsford2020).

Additional patterns observed in French have led other researchers to posit iambs (e.g., Charette Reference Charette1991; Weeda Reference Weeda1992; Armstrong Reference Armstrong1999; Goad & Buckley Reference Goad and Buckley2006). For example, an iambic analysis finds support from truncation – French has many disyllabic truncated forms, which exhibit final stress (Scullen Reference Scullen1997), as shown in (5).

In this way, French appears to be like Japanese, where there are a number of subminimal lexical words, yet truncation yields binary outputs motivating the foot (Itô Reference Itô1990). The evidence from truncation in French, however, is weakened by the presence of longer truncated forms (e.g., for paranoïaque ‘paranoid’ and for collaborateur/collaboratrice ‘collaborator’), as well as monosyllabic truncated forms (e.g., for triage ‘triage’ and for baccalauréat ‘baccalaureate’).

Although there is no broad consensus on the status of the foot in French, an important aspect of Selkirk’s (Reference Selkirk1978) analysis is that footing is not always associated with metrical prominence. Our analysis of HVD in QF will adopt this general idea: we will argue that HVD motivates the presence of a right-aligned iambic foot and, further, that footing is iterative, in the absence of stress. Our proposal builds on Verluyten (Reference Verluyten1982), who posits that HVD in QF is rhythmic: it preferentially applies in even-numbered syllables from the right word edge.

3. High vowel deletion in Québec French

In QF, there are a number of variable lenition processes that target high vowels.Footnote ¹ We compare devoicing and deletion, as researchers have often suggested that they might be one and the same process.Footnote ² Both are conditioned by linguistic and sociolinguistic factors (Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985); however, the linguistic factors do not substantially overlap, indicating that they are independent. Below, we introduce some of the constraints that regulate each process, as this will highlight the role of prosodic factors driving deletion in contrast to assimilatory factors driving devoicing.

Devoicing has a strict segmental conditioning: high vowels surface as devoiced between two voiceless consonants or either following or preceding a voiceless consonant (e.g., équiper ‘to equip’, but * robinet ‘tap’; Gendron Reference Gendron1966; Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985; Bayles Reference Bayles2016). Deletion, on the other hand, is not conditioned by voicing; high vowels may delete following or preceding both voiceless and voiced consonants (Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985; Bayles Reference Bayles2016; Garcia et al. Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017). Although the preceding consonant is an important predictor in HVD, it is manner of articulation that is most relevant: preceding obstruents favour the process, regardless of their voicing value (Dumas Reference Dumas1977; Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985; Cedergren Reference Cedergren and Sankoff1986).

Studies have also found that syllable shape affects HVD, in ways that suggest that deletion and devoicing are independent processes. Dumas (Reference Dumas1977) and Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) point out that HVD is preferred in CV syllables, and dispreferred in syllables with a complex onset or a coda. In addition, the shape of the consonant string that results from HVD seems to affect the likelihood of the process taking place. Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) note that HVD is avoided when it results in a long consonant string (such as in administration ‘administration’). These restrictions do not hold of high vowel devoicing; although devoicing seems to be dispreferred in syllables with complex onsets, we conjecture that this is not because of the complex onset structure per se, but rather because the second member of such clusters is a sonorant and thus voiced. High vowel devoicing also applies before voiceless codas, presumably because vowels are shorter in this context, an optimal circumstance for devoicing. In addition, the profile of the resulting consonant cluster may impact HVD, although in diverging ways across studies. Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017) observe that listeners judge HVD as dispreferred when the resulting consonant cluster mirrors a possible complex onset in QF (e.g., in soupirer ‘to sigh’, where is a possible onset cluster in the language), as deletion masks the original syllable structure. Lamontagne (Reference Lamontagne2023), in contrast, finds that production data favour deletion over retention of high vowels between consonants of this profile.

Turning to prosodic factors, devoicing may target high vowels in adjacent syllables, which is not possible with deletion (Walker Reference Walker1984); compare (6a) and (6b).

In addition, devoicing is possible word-finally when the PWd-final high vowel is not at the right edge of a PPh,Footnote ³ although it is less frequent in word-final position than it is word-medially (Walker Reference Walker1984). In contrast, many studies have shown that HVD is blocked at the right edge of a PWd or PPh (e.g., Dumas Reference Dumas1977; Verluyten Reference Verluyten1982; Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985; Couturier Reference Couturier1995), as can be seen in (7).Footnote ⁴

Finally, studies that have examined words in isolation have found that HVD is dispreferred in the first syllable of a word, as shown in (8), either in production (such as Couturier Reference Couturier1995), or based on speakers’ judgements (such as Guzzo et al. Reference Guzzo, Goad and Garcia2016). Couturier (Reference Couturier1995) points out that this dispreference may be connected to QF’s optional left-edge H tone, although he acknowledges that his data are insufficient to support this possibility.

Taken together, the data in (6)–(8) suggest that HVD is conditioned by word-level rhythmic structure: the process targets non-prominent syllables within the domain of the PWd. Accordingly, previous analyses (Verluyten Reference Verluyten1982; Cedergren & Simoneau Reference Cedergren, Simoneau, Lemieux and Cedergren1985; Cedergren Reference Cedergren and Sankoff1986; Garcia et al. Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017) have proposed that metrical structure constrains HVD. However, they do not agree on the kind of metrical structure that is assigned to the PWd level in QF, as we detail below.

Verluyten (Reference Verluyten1982), based on data obtained by Hammond (Reference Hammond1980), proposes that HVD preferentially applies in metrically weak syllables. In his proposal, a metrical rule maps the syllables in a given word to strong (S) and weak (W) nodes, which are assigned from right to left in alternating fashion. The resulting word-level rhythm is therefore iambic, the exception being words that end in a schwa, which is associated with a W node. Following from this, HVD is preferred in syllables associated with W, as shown in (9).

In contrast, Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) and Cedergren (Reference Cedergren and Sankoff1986) argue that HVD is not conditioned by alternating rhythm in QF. Instead, HVD is possible due to the inherent metrical profile of high vowels. In their analyses, prominence is represented with a metrical grid (as in Prince Reference Prince1983; Selkirk Reference Selkirk1984). First, all syllables are assigned a demi-beat (DB), as shown in (10). Subsequently, all non-final syllables that do not contain a schwa or a high vowel are assigned a basic beat (BB) at the immediately higher level. HVD, as well as schwa deletion, may apply to all syllables that are not aligned with a BB. This is exemplified in (10), where /i/ may be deleted in (10a) animal ‘animal’ and either /i/ or /y/ may be deleted in (10b) constitution ‘constitution’. The word-final syllable is assigned an additional beat in accordance with Cedergren’s (Reference Cedergren and Sankoff1986) Main Word Stress (MWS) rule – since word-final high vowels do not delete, they are assigned a beat at each level of the representation, like other (non-schwa) final syllables.

Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) and Cedergren (Reference Cedergren and Sankoff1986) consider HVD to be a result of variable phonetic effects that apply to vowels assigned only a DB in the metrical grid. However, in their analyses, Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) and Cedergren (Reference Cedergren and Sankoff1986) do not include any predictors related to word-level rhythmic profile. Thus, it is unclear whether there is any rhythmic conditioning in the sociolinguistic data they examine that is consistent with Verluyten’s (Reference Verluyten1982) proposal.

The idea that alternating rhythm conditions HVD in QF does find support in experimental data obtained by Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017). Native speakers of QF judged the naturalness of the pronunciation of words with and without deletion of /i/. Garcia et al. found that HVD is favoured in even-numbered syllables from the right word edge, suggesting iterative iambic footing, in the spirit of Verluyten (Reference Verluyten1982). In addition, they observed an interaction between metrical structure and morphology, as HVD is favoured when the target vowel is at the left edge of a suffix and in foot-dependent position (e.g., organiser ‘to organise’, exclusivité ‘exclusivity’).

It should be noted that, even though different studies have considered the role of metrical structure in HVD, only a subset include predictors relating to it, namely, Couturier (Reference Couturier1995), Guzzo et al. (Reference Guzzo, Goad and Garcia2016) and Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017). However, these studies differ in the metrical properties that were considered. On the one hand, Couturier (Reference Couturier1995) examined HVD in all positions in the word (including word-finally) and in words located in various phrasal positions, but while he considered predictors such as the number of syllables in the word, he did not include syllable structure. Guzzo et al. (Reference Guzzo, Goad and Garcia2016) and Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017), on the other hand, investigated HVD in words in isolation given predictors such as word-internal rhythm, number of syllables and (resulting) syllable structure, but while some of their items had multiple deletion sites, they did not probe which positions favoured deletion in such items. In other studies that have proposed a metrical approach for HVD, such as Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985), Cedergren (Reference Cedergren and Sankoff1986) and Verluyten (Reference Verluyten1982), predictors relating to metrical structure were not quantitatively examined. Although Cedergren & Simoneau (Reference Cedergren, Simoneau, Lemieux and Cedergren1985) and Cedergren (Reference Cedergren and Sankoff1986), in their analyses of sociolinguistic data, included several predictors relating to segmental profile, they did not control for metrical predictors, such as number of syllables in the word and position of the word in the phrase. In the case of Verluyten (Reference Verluyten1982), who examines sociolinguistic data first studied by Hammond (Reference Hammond1980), it is unclear whether any segmental or (additional) metrical predictors constrain the patterns that were observed.

In summary, while some previous studies have investigated the role of certain metrical constraints on HVD, they did not include other prosodic predictors that may also have an effect on the process. One such predictor is the left-edge H tone, which may hinder HVD when associated with a word-initial deletable vowel. Another issue to be considered is that, because these studies examined real words, phonotactic profile could not be strictly controlled for, which could mask some effects related to the metrical conditioning of HVD. We address these concerns in two experiments, which are detailed in the following sections.

4. Hypotheses

As is clear from the preceding, any conclusion arrived at regarding the prosodic conditioning of HVD is uncertain. The lack of consensus, we believe, stems principally from the finding that HVD applies variably. Consequently, linguistic and sociolinguistic factors can militate against deletion, even if the prosodic context is one that could favour deletion. Another consideration is that there may be competing prosodic constraints in play. Consider a word like limitation ‘limitation’ which, according to Verluyten, has the rhythmic profile WSWS. If HVD is sensitive to rhythm, the initial W syllable li should be the favoured site for HVD relative to the following S syllable mi; li, however, falls in the preferred location for the left-edge H tone if limitation is the first lexical word in a phrase. Thus, li could resist deletion relative to mi for this reason.

In order to tease apart these potentially competing prosodic factors, we report on two experiments, which test the role of each factor independently. We put forward two hypotheses as follows:

• Hypothesis 1: HVD is conditioned by the tonal profile of the PPh. HVD is dispreferred in the PWd-initial syllable of the leftmost PWd in a PPh, as this syllable is normally associated with a H tone.

• Hypothesis 2: HVD is conditioned by word-internal constituency. HVD is dispreferred in odd-numbered syllables from the right edge of the word, as these syllables fall into foot-head position in iambic feet.

Hypothesis 1 is addressed in Experiment 1, Hypothesis 2 in Experiment 2. Specific predictions that stem from each hypothesis are discussed in §§5.2 and 6.2.

5. Experiment 1: HVD and phrasal prominence

To our knowledge, virtually all previous research on HVD has examined words in isolation. The one exception we have found is Couturier (Reference Couturier1995), who looked at words both in isolation and in various positions in sentences and found that they displayed comparable behaviour with respect to HVD.Footnote ⁵ The restriction to isolated words stems from the observation that HVD does not cross PWd boundaries. For example, in a phrase like joli sens ‘beautiful meaning’, PWd-final is not a candidate for deletion, although the same vowel is deletable in the segmentally quasi-parallel phrase la licence ‘the licence’. Because previous research has observed that HVD is delimited by the PWd, any potential role that the left-edge H tone may play in regulating the process has not been examined, given that it is assigned as part of the tonal profile of the PPh. We aim to fill this gap with Experiment 1.Footnote ⁶

Recall from §2 that prominence assignment in QF occurs at the level of the PPh. It is thus possible that lenition phenomena, such as HVD, are constrained by the prosodic profile of the PPh rather than the PWd. In Experiment 1, we examine this possibility. Specifically, we employ a judgement task using auditory stimuli to investigate whether QF speakers’ preferences regarding HVD are influenced by the phrase-level tonal profile, where, in addition to the obligatory H tone at the end of the PPh, another H tone is associated with the left edge of the first lexical word in the domain (as shown in (2)). If HVD is constrained by phrase-level prominence, then it should be avoided in syllables that would normally receive a H tone. To illustrate, word-initial HVD should be rated worse overall in determiner + noun strings (e.g., la mythologie ‘the mythology’), because the deletable vowel (underlined) is in the first syllable of the noun which, as the first (and only) PWd in the phrase, is typically assigned an initial H tone. In contrast, HVD should be rated better in determiner + adjective + noun structures (e.g., la jolie mythologie ‘the beautiful mythology’), because the deletable vowel is not assigned a H tone, as the initial H tone appears on the first syllable of the preceding adjective. This experiment is detailed in the sections that follow.

5.3. Results

Figure 1 shows participants’ judgements of items with HVD by number of syllables (two vs. four) and type of phrase (N, DN or DAN). As previously mentioned, participants rated the items based on a 4-point scale, where 1 corresponded to ‘completely unnatural’ and 4 corresponded to ‘completely natural’. The patterns in Figure 1 suggest that HVD is overall more natural in four-syllable nouns, regardless of the type of phrase.

Figure 1

Participants’ judgements of items with HVD in the word-initial syllable by number of syllables and type of phrase, where 1 (white) corresponds to ‘completely unnatural’ and 4 (dark grey) corresponds to ‘completely natural’.

Figure 1 Long description

The chart has two main vertical sections: Number of syllables 2 on the left and Number of syllables 4 on the right. Each section contains three horizontal bars labeled D A N, D N, and N, representing phrase types. Each bar is divided into four segments, corresponding to ratings 1 to 4, from left to right. For two syllables: D A N shows 40 percent for rating 1, 29 percent for rating 2, 16 percent for rating 3, and 16 percent for rating 4. D N shows 45 percent for rating 1, 21 percent for rating 2, 17 percent for rating 3, and 18 percent for rating 4. N shows 52 percent for rating 1, 16 percent for rating 2, 14 percent for rating 3, and 18 percent for rating 4. For four syllables: D A N shows 14 percent for rating 1, 19 percent for rating 2, 27 percent for rating 3, and 40 percent for rating 4. D N shows 21 percent for rating 1, 20 percent for rating 2, 24 percent for rating 3, and 34 percent for rating 4. N shows 26 percent for rating 1, 19 percent for rating 2, 21 percent for rating 3, and 33 percent for rating 4. Higher ratings (3 and 4) are more frequent for four-syllable items across all phrase types.

The data were analysed using a Bayesian hierarchical ordinal regression using the brms package (Bürkner Reference Bürkner2017) in R (R Core Team 2025) with default priors (flat) for main effects and for thresholds (student t). The model included type of phrase and number of syllables as main effects, in addition to by-participant random slopes (for both variables) and intercepts and by-item random intercepts. The model in question was run using 8,000 iterations, 4 chains and 2,000 warm-up steps, which guaranteed a satisfactory effective sample size of at least 4,500. Finally, our chains were visually inspected and our R̂ values were checked to ensure that the model had converged. We ran models with and without an interaction between type of phrase and number of syllables. We compared the models using their widely applicable information criterion (WAIC) values (Watanabe Reference Watanabe2013); as the two models performed similarly, we report the model without interaction.

Figure 2 shows the posterior distributions of credible effect sizes for each predictor variable. On the right, the mean effect size for each posterior is displayed, along with its 95% credible interval (CrI; in the distributions that follow, which are all approximately Gaussian, this is synonymous with the highest density interval (HDI)). Simply put, parameter values closer to the peak of the distribution are more likely given the data than parameter values that are distant from the peak. Effects must be interpreted relative to phrases of the form N with two-syllable nouns, which represent our reference levels. Finally, Figure 2 also displays the region of practical equivalence (ROPE) around zero. If an entire CrI were to be found within this region, we would conclude that a null effect has been found. This is not the case here. As we can see in the figure, words with four syllables favoured a higher rating by participants relative to words with two syllables, as indicated by the positive posterior distribution in the figure. Neither type of phrase (DN or DAN) shows a statistically credible effect relative to N in our model.

Figure 2

Posterior distributions of effect sizes with associated means and 95% credible intervals (CrIs). Shaded area surrounding zero represents the region of practical equivalence (ROPE).

Figure 2 Long description

From top to bottom, three horizontal density curves represent Syll_4, D N, and D A N. The x-axis is labeled posterior distribution, spanning negative two to four. Each curve is shaded, with a vertical dashed line at zero and a shaded region around zero marking the region of practical equivalence. Syll_4 is the top curve, centered right of zero, with mean 1.53 and credible interval 0.76 to 2.29. D N is the middle curve, centered near zero, with mean 0.19 and credible interval negative 0.72 to 1.12. D A N is the bottom curve, also near zero but slightly right-shifted, with mean 0.53 and credible interval negative 0.38 to 1.43. All means and intervals are shown to the right of each curve.

These results do not confirm our predictions about the role of the left-edge H tone on HVD, since, as shown in Figure 1, all types of phrases for two-syllable nouns and all types of phrases for four-syllable nouns have similar ratings. Furthermore, participants’ preferences are not in the direction we predicted, as HVD is overall preferred in longer nouns regardless of phrase type and, in consequence, regardless of whether the deletable vowel may be associated with a H tone.

6. Experiment 2: HVD and word-internal constituency

As mentioned, previous research has found mixed results with respect to the role of metrical structure in HVD in QF. Verluyten’s (Reference Verluyten1982) proposal that HVD is favoured in even-numbered syllables from the right edge of the word, consistent with iambic footing, was later challenged by Cedergren (Reference Cedergren and Sankoff1986) based on sociolinguistic data, but supported by Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017) in a judgement task with auditory stimuli. Although the study reported in Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017) controlled for the syllabic profile of the target words (i.e., target /i/ was always in a non-final open syllable and was preceded by a singleton, non-affricated onset), it included a number of additional potential predictors, such as resulting phonotactic profile and morphological status of the deletable vowel (i.e., in the stem or in a suffix), which may conceal or amplify the effects of metrical structure on the process. In addition, Garcia et al. (Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017) had participants judge individual words with or without deletion, and although some of the items included in their experiment had more than one deletable vowel, they were not analysed separately to probe the positions in which HVD would be preferred. As a result, it is impossible to determine whether HVD in a given position in the word would be judged to be better relative to HVD in another position in an item with a similar shape. To address these concerns, and to further explore the possibility that HVD is constrained by the head foot domain, we conducted a judgement task where participants had to choose between two alternative pronunciations of nonce words, which were designed to follow a strict phonotactic template.

6.1. Methods

6.1.1. Participants

Participants were 39 native speakers of QF, all of whom grew up in the Greater Montréal area or surrounding regions.Footnote ⁸ Some of the participants reported having knowledge of other languages (including English) at various proficiency levels, but none were bilingual from childhood. The mean age of participants was 27.3, ranging from 18 to 63. None of the participants reported any hearing impairments.

6.1.2. Stimuli

The target stimuli were nonce words with four, five or six syllables. The target words had two high vowels (always /i/) and were recorded in two ways, each time with deletion of one of the high vowels. In four-syllable words, the deletable vowel was in syllable position 2 or 3 (counting from the right edge of the word). In some five-syllable words, the deletable vowel was in position 2 or 4, while in others, it was in position 3 or 4. In six-syllable words, the deletable vowel was located in position 4 or 5. In contrast to the stimuli in Experiment 1, the initial syllable was never targeted for deletion, as our objective was to examine whether HVD is conditioned by alternating rhythm, without the influence of any possible predictors associated with word edges. Forty-eight nonce words were assigned to each of the four conditions, totalling 192 target items. A sample of the target stimuli is given in Table 1 (a full list is provided in Appendix B).

Table 1

Examples of target items in Experiment 2 across four conditions, varying by number of syllables and position of deletion. The deletable vowels are underlined

Table 1 Long description

Starting at the top-left, the first column lists four-syllable words with deletion sites at syllable 2 or 3, such as dab i n i bo, kan i b i nɛ, fɛɡ i v i ɡa, and ʁɔv i ɡ i va, each with two underlined i vowels. The second column presents five-syllable words with deletion sites at syllable 2 or 4, including kɛb i nab i no, ʃɔn i ban i be, lɔɡ i vaɡ i ve, and ʁav i ɡɛv i ɡo, each with two underlined i vowels. The third column contains five-syllable words with deletion sites at syllable 3 or 4, such as tab i n i bɔfɛ, ʃɔn i b i neʁo, tɛɡ i v i ɡɔʁa, and lav i ɡ i vɔʁɛ, again with two underlined i vowels per word. The fourth column displays six-syllable words with deletion sites at syllable 4 or 5, including kɔb i n i beʃɔʁa, pɛn i b i nɔfɛpa, ʁaɡ i v i ɡɔʒɛle, and tɛv i ɡ i vɔʒape, each with two underlined i vowels. Each cell contains a phonetic transcription in square brackets, and underlining consistently marks the vowels eligible for deletion according to the experimental condition.

The syllables in the target stimuli were invariably CV, and the target vowels were always preceded and followed by voiced consonants, to avoid judgements based on the perception of vowel devoicing instead of deletion. In addition to target /i/, the stimuli included the vowels . Other high vowels, as well as nasal vowels and schwa, were avoided. In target items with four and six syllables, as well as in five-syllable items where deletion was in either position 3 or position 4, the deletable vowels were in adjacent syllables.

The portions of the target stimuli containing the deletable vowels were built based on specific templates, to avoid judgements that could stem from consonant quality rather than position in the word. For example, in four-syllable words, template alternated with , which resulted in the deletable vowels (in syllable 2 or 3) being placed in identical phonotactic contexts (e.g., , ). In five-syllable items with deletion in position 2 or 4, the deletable vowels were preceded and followed by the same segments (e.g., , ). As these examples indicate, each template was duplicated such that the consonants were in the inverse order across pairs of stimuli (e.g., $\sim$ , $\sim$ ). Each template was repeated (with different preceding and following segments) three times per condition.

The fillers were designed based on the target stimuli, by replacing one of the target high vowels with /a/. To avoid adjacent low vowels in some of the fillers, an /a/ that was present in the original target item was replaced by another non-high vowel (). The total number of fillers was also 192.

The stimuli were recorded by a male native speaker of QF from the Montréal area with training in linguistics and phonetics. Parallel to Experiment 1, no left-edge H tone was produced in the test items, to prevent participants from making judgements based on the surface tonal profile. The stimuli were checked in Praat (Boersma & Weenink Reference Boersma and Weenink2024) to ensure that no traces of the deleted vowel were present.

6.1.3. Procedure

The experiment was developed in OpenSesame (Mathôt et al. Reference Mathôt, Schreij and Theeuwes2012). Participants were presented with two auditory versions of the same word in sequence, each with deletion of a different vowel (e.g., vs. for the item ). There was a pause of one second between the two stimuli. While participants were listening to both stimuli, the corresponding orthographic form was displayed on the screen. The orthographic forms of the test items were determined following the sound-to-letter correspondence conventions observed in French and were checked by a native speaker of QF. Participants were told that they would be listening to (and reading) new words introduced into French. They were asked to decide which of the two pronunciations that they heard for each new word sounded more natural.

There were three versions of the task. Each version started with a practice block containing six filler items (identical across versions), to familiarise participants with the procedure. The 192 experimental items were equally divided across the three versions (64 each). Each version of the experiment also included all the remaining fillers (i.e., 186), for a total of 256 stimuli per version.Footnote ⁹ The stimuli were presented in three blocks. Participants were encouraged to take a short break between blocks.

The experiment took approximately 45 minutes. Participants were tested in a sound-attenuated booth at McGill University and were compensated for their time.

6.2. Predictions

For Experiment 2, we hypothesised that HVD would be preferred in foot-dependent position over foot-head position. This hypothesis was based on previous findings supporting the metrical conditioning of HVD (Verluyten Reference Verluyten1982; Garcia et al. Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017). Although Experiment 1 was designed to test a different hypothesis – that HVD is modulated by the left-edge H tone of the French prominence system – the experiment instead yielded results that are consistent with Hypothesis 2: assuming that deletion is regulated by metrical structure, it was found to be dispreferred in the dependent position of the head foot. However, an alternative interpretation of the findings was also available, that deletion is more common in longer words. The four types of stimuli included in Experiment 2 were designed to further probe this issue. Our predictions are as follows.

If HVD is sensitive to iteratively built iambic feet, we predict that deletion should be more common in the three types of stimuli where the vowel targeted for deletion is in foot-dependent position as compared to foot-head position; that is, deletion in position 2 in four-syllable words, relative to position 3; and deletion in position 4 in five- and six-syllable words, relative to positions 3 and 5, respectively. This is expressed in Prediction 2.1, where A > B indicates that deletion in strings of shape A is predicted to show a statistically credible effect relative to deletion in strings of shape B (bold marks the foot head; underlining marks the location of the deleted vowel).Footnote ¹⁰

• Prediction 2.1: HVD in foot-dependent position will be favoured over HVD in foot-head position.

  $ \kern-12pt \begin {array}{lr@{~>~}r} \text{ Four-syllable words:} & ({\sigma}\boldsymbol{\sigma})({\underline {\sigma}\boldsymbol {\sigma}}) \gt & ({\sigma {\underline {\sigma}}})({\sigma\boldsymbol {\sigma}}) \\ \text{ Five-syllable words (deletion in position 4 or 3):} & {{\sigma}(\underline {\sigma}\boldsymbol {\sigma})(\sigma\boldsymbol {\sigma})} \gt & {\sigma(\sigma {\underline {\sigma}})(\sigma\boldsymbol {\sigma})} \\ \text{ Six-syllable words:} & {(\sigma\boldsymbol {\sigma})(\underline {\sigma}\boldsymbol {\sigma})(\sigma\boldsymbol {\sigma})} \gt & {(\sigma{\underline {\sigma}})(\sigma\boldsymbol {\sigma})(\sigma\boldsymbol {\sigma})} \end {array}$

If French builds iambic feet across the word, one foot in the PWd will be the head foot. As French displays final prominence, it is the rightmost foot that will be marked as the head foot. In Experiment 1, we found that HVD is preferred in the dependent position of the non-final foot in four-syllable words compared to the dependent position of the only foot in two-syllable words. An interpretation of this finding is that deletion is disfavoured in the head foot. In Experiment 2, one comparison enables us to further test this interpretation.

For five-syllable words with HVD in position 2 or 4, both deletion sites are in foot-dependent position, but the former is in the head foot while the latter is not. Thus, if the head foot protects vowels from deletion, then we predict that HVD will be favoured when the vowel targeted for deletion is outside this foot.

• Prediction 2.2: HVD will be favoured when the vowel targeted for deletion is outside the head foot.

  $\kern-12pt \begin {array}{lr@{~>~}r} \text{ Five-syllable words (deletion in position 4 or 2):} & {\sigma(\underline {\sigma}\boldsymbol {\sigma})(\sigma\boldsymbol {\sigma})} \gt & {\sigma(\sigma\boldsymbol {\sigma})(\underline {\sigma}\boldsymbol {\sigma})} \end {array}$

Table 2 summarises our predictions for each condition in Experiment 2. We report our results in §6.3.

Table 2

Predictions for Experiment 2 based on number of syllables in stimuli and position of deletable vowel (underlined)

Table 2 Long description

Starting from the top row, the table lists four cases. First row: four syllables, deletable positions three or two, H V D preferred in position two with dabin i bo, dispreferred in position three with dab i nibo. Second row: five syllables, deletable positions four or three, H V D preferred in position four with tab i nibɔfɛ, dispreferred in position three with tabin i bɔfɛ. Third row: six syllables, deletable positions five or four, H V D preferred in position four with kɔbin i bɛʃɔʁa, dispreferred in position five with kɔb i nibɛʃɔʁa. Fourth row: five syllables, deletable positions four or two, H V D preferred in position four with kɛb i nabino, dispreferred in position two with kɛbinab i no. Each stimulus shows the deletable vowel underlined within the phonetic sequence.

6.3. Results

Figure 3 shows participants’ preferences for HVD by number of syllables in the item and position of the deleted high vowel. The figure suggests that, in four- and six-syllable items, participants preferred HVD in foot-dependent positions, that is, in positions 2 and 4, respectively. On the other hand, in five-syllable items, positions 3 and 4 (second panel) and positions 2 and 4 (rightmost panel) were chosen as the preferred sites for HVD at very similar rates.

Figure 3

Participants’ preferences for HVD by number of syllables and position of deletable vowel.

Figure 3 Long description

The graph contains four groups of bars, each labeled above with syllable length and deletable sites. From left to right, the first group is labeled Length 4-syll, Deletable sites 3–2, with bars for positions 3 and 2. The bar for position 2 is higher, reaching just above 60 percent, while position 3 is near 45 percent. The second group is Length 5-syll, Deletable sites 4–3, with bars for positions 4 and 3; position 3 is higher, near 55 percent, and position 4 is near 45 percent. The third group is Length 6-syll, Deletable sites 5–4, with bars for positions 5 and 4; position 4 is higher, just above 60 percent, and position 5 is near 45 percent. The fourth group is Length 5-syll, Deletable sites 4–2, with bars for positions 4 and 2; position 2 is higher, near 55 percent, and position 4 is near 45 percent. All bars include error bars. The y-axis is labeled Preference, ranging from 0 percent to 100 percent. The x-axis is labeled Position of deletion in stimulus. Across all groups, preference is consistently higher for the earlier deletable position.

To analyse these results, we ran two Bayesian logistic regressions, one for the data in the first three panels and one for the data in the last panel. The first regression models participants’ preferences for HVD in foot-head vs. foot-dependent position, to test Prediction 2.1. The second regression models participants’ preferences for HVD in the head foot vs. in a non-head foot, to test Prediction 2.2.

We start by discussing the first model. In this regression, participants’ preferences (foot-head position=1 vs. foot-dependent position=0) are modelled as a function of number of syllables. By-participant random intercepts and slopes (number of syllables) were also included, as were by-item random intercepts. The results of this model can be examined in Figure 4. As we can see, both four- and six-syllable words have entirely negative 95% CrIs. Negative effects indicate a dispreference for stimuli where deletion occurs in foot-head position relative to five-syllable words, our intercept (5-syll*). For five-syllable words, in contrast, the posterior distribution includes zero in its 95% CrI, which indicates that zero log-odds (i.e., a probability of 50%) is a statistically credible effect given the data. Overall, the posterior distributions in Figure 4 confirm the trend observed in Figure 3 for both four- and six-syllable words. Finally, no credible difference exists between four- and six-syllable words, which is unsurprising given the patterns observed in Figure 3 and their respective posterior distributions shown in Figure 4. These results partially support Prediction 2.1, as HVD is preferred in foot-dependent position in two of the three conditions under analysis.

Figure 4

Posterior distributions of effect sizes (in log-odds) with associated means and 95% credible intervals (CrIs). Shaded area surrounding zero represents the region of practical equivalence (ROPE). Both distributions must be interpreted as differences relative to the intercept (five-syllable words, represented with an asterisk in the figure).

Figure 4 Long description

The horizontal density plot has three stacked distributions. The x-axis is labeled Posterior distribution, ranging from negative 2 to 2. The topmost curve is labeled 5-syll, marked with an asterisk, and centered near zero, with a mean of 0.16 and a 95 percent credible interval from negative 0.15 to 0.47. The middle curve is labeled 6-syll, with a mean of negative 0.53 and a 95 percent credible interval from negative 1.04 to negative 0.02. The bottom curve is labeled 4-syll, with a mean of negative 0.7 and a 95 percent credible interval from negative 1.29 to negative 0.12. A vertical dashed line at zero marks the reference value, and a shaded region around zero indicates the region of practical equivalence. All distributions are interpreted relative to the 5-syllable reference group.

Returning to Figure 3, the final panel suggests that no credible difference exists between positions 2 and 4 in five-syllable words. We ran a second Bayesian logistic regression to confirm if this is the case. This second regression is an intercept-only model with by-participant and by-item random intercepts including only five-syllable words with deletion in position 2 or 4; recall that both are in foot-dependent position but differ as to whether they are in the head foot (position 2) or in a non-head foot (position 4). The posterior distribution comparing deletion in both positions supports the observation that participants’ responses for such positions are not statistically different: $\hat {\beta}$ = 0.19, 95% CrI = [−0.18, 0.57]. This result does not support Prediction 2.2, as zero is a relatively credible parameter value for the intercept in our model (between the lower threshold of our 95% CrI and the peak of the posterior distribution for the intercept).

In summary, a robust effect was found for both four- and six-syllable words whereby participants disfavour deletion in foot-head position, which is partially consistent with Prediction 2.1. Contrary to Prediction 2.1, however, this same finding was not observed for five-syllable words, as these words show similar preferences for deletion in both positions under investigation. In addition, there is no statistically credible evidence in the data that participants disprefer HVD in the head foot of the word, contrary to Prediction 2.2. We discuss these findings in the next section.

7. General discussion and conclusion

We conducted two experiments to investigate whether HVD in QF is conditioned by prosodic factors. In Experiment 1, we tested Hypothesis 1, that HVD is sensitive to the tonal profile of the PPh. Specifically, we examined whether HVD in initial position is dispreferred in phrasal contexts where the deletable vowel is associated with a H tone. In Experiment 2, we tested Hypothesis 2, that HVD is regulated by word-internal constituency. We probed whether deletion is dispreferred in foot-head position as well as in the head foot of the PWd.

Our results for Experiment 1 indicate that word size rather than type of phrase constrains HVD, as participants dispreferred deletion in all types of phrases with two-syllable nouns. In other words, deletion was preferred in initial position in longer nouns (with four syllables), regardless of type of phrase (N, DN or DAN). These results contradicted our predictions, according to which HVD would be dispreferred in contexts where the initial syllable of the noun could be associated with a H tone (i.e., in four-syllable nouns in N and DN phrases), and thus Hypothesis 1 was not supported. We considered two possible explanations for participants’ behaviour, namely, that HVD is dispreferred in two-syllable words because deletion in shorter words could compromise lexical retrieval; and that HVD is dispreferred in two-syllable words because the only foot in the PWd where the deletable vowel could be located is the head foot.

We tested the latter possibility in Experiment 2, as part of the hypothesis that HVD is overall conditioned by word-internal constituency. The results for Experiment 2 do not confirm the prediction that HVD is dispreferred in the head foot of the PWd, as participants rated the naturalness of five-syllable target words with deletion in position 2 or 4 very similarly. Nonetheless, as both syllables in such words could be in foot-dependent position, these results are in line with the hypothesis that HVD is conditioned by word-internal constituency. Regarding the other items included in Experiment 2, participants disfavoured HVD in foot-head relative to foot-dependent position, as predicted, but this dispreference was observed only in four- and six-syllable words, not in five-syllable words.

Taken together, the results for Experiments 1 and 2 suggest that HVD in QF is prosodically conditioned. This conditioning stems from PWd-internal metrical structure, and not from the assignment of PPh-level prominence. The question we must address is what type of metrical structure holds in QF to capture the deletion patterns we have observed. The patterns for four- and six-syllable words in Experiment 2 suggest that HVD is sensitive to alternating rhythm, which we formally express using feet.

We explore four options for the construction of feet. The first three options, which we show do not work, involve multiple feet for all word shapes under consideration: multiple iambs, multiple trochees or both iambs and trochees. These options largely cover the range of what we would expect for French based on earlier literature. The last option differs from the others in that it proposes distinct parsings for even- and odd-parity forms.

Option 1 is in Table 3: disyllabic iambs are iteratively built from the right edge of the word, as shown under the column entitled expected deletion preferences. Option 1 predicts HVD to be preferred in even-numbered syllables from the right word edge, that is, in foot-dependent position; and that this preference extends beyond the rightmost foot, because footing is iterative, even though there is no evidence for this from stress (heads of binary feet are in bold in all following tables).

Table 3

Option 1: Iterative iambic footing

Table 3 Long description

The table has four rows, each anchored by the number of syllables in the first column: 4, 6, 5, and 5. The second column lists H V D site values: 3 or 2, 5 or 4, 4 or 3, and 4 or 2. The next three columns under Expected deletion preferences display forms with bold and underlined segments, such as (da b i)(n i bo) and (kɔ b i)(n i be)(ʃɔ ʁa). The following three columns under Attested patterns show actual forms, including dabin i bo, kɔbin i beʃɔʁa, tabin i bɔfɛ, and kɛbinab i no, with underlining and bolding marking specific segments. Symbols such as question marks and equals signs appear in the fourth and seventh columns, indicating uncertainty or equivalence between forms. Each row aligns expected and attested patterns for a given syllable count, highlighting differences in segment deletion and preservation.

For four- and six-syllable words, as well as for five-syllable words with deletion in position 4 or 2, the representations that result from this parse capture the findings from Experiment 2, as shown by a match between the symbols (> or =) in the expected deletion preference and attested pattern columns. In the former column, A > B indicates that deletion in strings of shape A is preferred over strings of shape B for the foot parse under consideration, whereas A = B indicates no preference; in the latter column, A > B indicates that deletion in strings of shape A showed a statistically credible effect relative to deletion in strings of shape B, whereas A = B indicates the absence of a credible difference.

In contrast to the three word shapes just discussed, for five-syllable words with deletion in position 4 or 3, iteratively built iambs fail to capture the experimental findings: the expected deletion preference is the same as that observed for four- and six-syllable words, yet the attested pattern is that where HVD is equally favoured in foot-dependent and foot-head positions. Indeed, it appears that for five-syllable words, and perhaps for all odd-parity words, HVD is not sensitive to iterative footing.

In view of this, we turn to Option 2, which assumes that HVD is regulated by trochees iteratively built from the right word edge. Recall that Selkirk (Reference Selkirk1978) proposes that all syllables with a full vowel or a coda form a foot in French, while syllables containing schwa, which is prosodically deficient, are parsed with the preceding syllable, forming a trochee.

This analysis for schwa was extended to high vowels by Bosworth (Reference Bosworth2017) to capture the observation that high vowels, like schwa, can be weakened, undergoing laxing or deletion in QF. As weakening can be observed in various non-final positions within the PWd, Bosworth proposes that non-final syllables with high vowels must be footed with another syllable, optimally with the preceding syllable (e.g., camisole ‘camisole’), alternatively with the following syllable, for example, when in word-initial position (e.g., numéro ‘number’). Adjacent syllables containing high vowels form a trochaic foot, expanding on Montreuil’s (Reference Montreuil, Agwuele and Park2002) analysis for schwa.

Bosworth (Reference Bosworth2017) predicts that word-internal high vowel weakening can take place in foot-head or foot-dependent position, but she proposes that it is preferred in the latter position. The foot parses and expected deletion preferences in Table 4 are consistent with her assumptions.Footnote ¹¹

Table 4

Option 2: Iterative trochaic footing

Table 4 Long description

The table has eight columns and four rows. The first column lists number of syllables: 4, 6, 5, and 5. The second column lists H V D site: 3 or 2, 5 or 4, 4 or 3, and 4 or 2. The third to fifth columns show expected deletion preferences, with examples such as (da)(bi n i)(bo), (kɔ)(bi n i)(be)(ʃɔ)(ʁa), (ta)(b i n i)(bɔ)(fɛ), and (kɛ b i)(na bi)(no). The sixth to eighth columns show attested patterns, with examples such as dabinibo, kɔbinibeʃɔʁa, tabinibɔfɛ, and kɛbinabino. Bold and underlined letters indicate the deleted or retained syllable. Symbols such as inverted question mark, exclamation mark, and equals sign appear in the fourth and seventh columns to indicate pattern status.

As in Option 1, four- and six-syllable words are correctly expected to pattern together under iterative trochaic footing. Option 2 also fails to correctly capture five-syllable words with deletion in position 4 or 3; in this case, the expected pattern is the inverse of that under Option 1.

We thus turn to consider Option 3, that footing reflects the ‘accentual arc’ or ‘hammock’ pattern of prominence observed in French (Fónagy Reference Fónagy, Fónagy and Léon1979). In phrases containing one PWd, a trochee would be located at the left edge of the PWd, the head of which is signalled by the left-edge phrasal H tone, and an iamb would be located at the right edge of the PWd, the head of which is signalled by final prominence (Goad & Buckley Reference Goad and Buckley2006).

Option 3 captures the observation that HVD is rarely observed in word-initial and -final positions, which would always be heads of feet under this analysis. It also predicts deletion to be equally preferred in all word-internal syllables. In so doing, it correctly predicts all five-syllable words to pattern together, regardless of deletion site, as shown in Table 5, but it fails to capture the finding that four- and six-syllable words, and perhaps all even-parity words, prefer deletion in even-numbered syllables from the right word edge. Finally, recall that the results from Experiment 1 do not support the idea that the initial H tone is associated with the head of a foot, since HVD is not conditioned by tonal profile. We thus reject the possibility that initial prominence in the accentual arc corresponds to a foot head and return to the assumption that feet in QF are built from the right edge only, as under Options 1 and 2.

Table 5

Option 3: Footing reflects accentual arc

Table 5 Long description

The table has four rows, each corresponding to a different number of syllables and H V D site. The first column lists the number of syllables as 4, 6, 5, and 5. The second column lists H V D site as 3 or 2, 5 or 4, 4 or 3, and 4 or 2. The next three columns under expected deletion preferences show bracketed and bolded or underlined syllables, such as open parenthesis bold d a close parenthesis b i open parenthesis n underlined i bold b o close parenthesis, with equals signs separating alternative forms. The next three columns under attested patterns show unbracketed forms with underlined vowels, such as d a b i n underlined i b o, and alternative forms separated by equals signs or question marks. Each row presents a different pattern of syllable deletion and realization, with the expected and attested forms aligned for comparison.

Both iambic and trochaic analyses (Options 1 and 2) appropriately predict four- and six-syllable strings to pattern as if they are sensitive to the contrast between foot head and foot dependent. However, no particular preference for five-syllable words with deletion in position 3 or 4 was observed, which cannot be captured under either of these options. The same lack of preference holds for five-syllable words with deletion in position 2 or 4, although in this case, none was expected and this was correctly captured by both footing options.

We suggest that the way forward is for both types of five-syllable words, regardless of deletion site, to be examined through the same lens, in contrast to four- and six-syllable words. More concretely, we propose that even- and odd-parity strings be parsed distinctly, as shown in Option 4 in Table 6. Even-parity forms are exhaustively parsed, as every syllable can be contained within a binary iambic foot. Odd-parity forms cannot be exhaustively parsed. Because one syllable is inevitably left unparsed in such forms (unless degenerate feet are accepted), we assume that foot construction stops once the head foot is built at the right edge of the word. It follows that HVD should be preferred in foot-dependent positions in even-parity forms, but that no preference regarding deletion site should be observed for HVD in odd-parity forms.Footnote ¹²

Table 6

Option 4: Iambic footing, iterative in even-parity strings, non-iterative in odd-parity strings

Table 6 Long description

Starting from the top row, the table lists four cases by syllable count: 4, 6, 5 (twice). The first column shows ‘No. of σs’ with values 4, 6, 5, and 5. The second column, ‘H V D site’, lists 3 or 2, 5 or 4, 4 or 3, and 4 or 2. The third column, ‘Expected deletion preferences’, displays bracketed syllable groupings with bold and underlined segments, such as (da b i)(n i bo) for 4 syllables, (kɔ b i)(n i be)(ʃɔ ʁa) for 6 syllables, tab i ni(bɔ fɛ) for 5 syllables, and kɛb i na(bi no) for 5 syllables. The fourth column contains symbols: inverted question marks for the first two rows and equals signs for the last two. The fifth column, also ‘Expected deletion preferences’, shows bracketed forms with bold and underlined segments, e.g., (da b i)(ni bo), (kɔ b i)(ni be)(ʃɔ ʁa), tabin i (bɔ fɛ), and kɛbina(b i no). The sixth column, ‘Attested patterns’, presents unbracketed forms with underlined segments: dabin i bo, kɔbin i beʃɔʁa, tab i nibɔfɛ, and kɛb i nabino. The seventh column repeats the fourth column's symbols. The eighth column, ‘Attested patterns’, lists dab i nibo, kɔb i nibeʃɔʁa, tabin i bɔfɛ, and kɛbinab i no. Bold and underlined formatting highlights specific syllables in each form.

Our analysis selects iambs over trochees for the following reason. An analysis based on trochees, as applied to French, assumes binary constituency only when high vowels (or schwas) are present. Similar to the analysis in Cedergren (Reference Cedergren and Sankoff1986), where all high vowels are associated only with a DB and are therefore susceptible to deletion (see (10)), a trochaic analysis depends on the position of high vowels in a word for feet to be formed. As a result, a trochaic analysis is unable to make predictions about the application of weakening processes, since all weak segments (such as high vowels and schwas) are equally susceptible to such processes, regardless of their position within the PWd.

Our analysis captures the attested patterns for HVD, but it rests on two assumptions that could be challenged: (i) QF requires distinct parsing mechanisms in even- and odd-parity words;Footnote ¹³ and (ii) QF builds feet iteratively in even-parity words, even though there are no prosodic cues to word-internal prominence outside of the final foot. In the following paragraphs, we motivate these assumptions more broadly.

The analysis we provided for QF in Table 6 entails that even- and odd-parity words involve different parameter settings (or constraint rankings) for foot construction. Specifically, the iterativity parameter must be set differently for words of different shapes. Since the number of syllables in a word depends on factors such as its morphological composition, the difference in footing cannot be stored in the lexicon instead.

We suggest that there is nothing unusual about QF in this respect, as distinct parses are required for other aspects of stress computation across languages. In Yidiɲ, for example, even- and odd-parity strings require different foot shapes. Footing is iambic in odd-parity forms and trochaic in even-parity forms (Dixon Reference Dixon1977; Hyde Reference Hyde2012), as shown in (15a) and (15b), respectively (examples adapted from Dixon Reference Dixon1977: 4–5 and Hyde Reference Hyde2012: 410). The difference in foot shape is supported by lengthening effects: lengthening targets the head foot in iambic forms only; trochaic forms do not display lengthening, consistent with cross-linguistic preferences (see Hayes Reference Hayes1985, Reference Hayes1995). The patterns observed in Yidiɲ are thus consistent with the idea that the parser must know how many syllables a word has before feet are built.

Wargamay presents a similar situation to Yidiɲ. In Wargamay, where stress is trochaic, the peninitial syllable of odd-parity forms (where the head foot is) is lengthened, but no lengthening is observed in even-parity forms; see (16) (examples adapted from Dixon Reference Dixon, Dixon and Blake1981: 20 and Hyde Reference Hyde2012: 411). In this case, for lengthening to apply, the parser must know if the head foot is left-aligned, which depends on the word having an even or odd number of syllables.

Concerning assumption (ii), we have proposed that QF builds feet iteratively in even-parity words, even though there are no prosodic cues to word-internal prominence outside of the final foot. Seminole/Creek resembles QF in this respect: feet must be built iteratively, but as it is a pitch-accent language, only the head foot receives prominence (realised as a H tone; Haas Reference Haas1977). Our analysis follows Hayes (Reference Hayes1995). In words where all syllables can be parsed into binary right-headed feet, the H tone falls on the final syllable, as shown in (17a). In words where this syllable cannot be parsed into a binary foot, the H tone instead falls on the penult, (17b). To obtain this difference in the location of the H tone, feet must be built iteratively from the left edge, even though the heads of non-final feet are not cued in any way.

In short, Seminole/Creek is similar to QF in that feet must be built iteratively so that a phonological process may correctly apply: H tone assignment in Seminole/Creek, HVD in even-parity forms in QF.

In summary, we propose that HVD in QF is sensitive to alternating iambic rhythm, but only in even-parity forms, which are fully parsed. In odd-parity forms, where feet are not built exhaustively, no preference between deletion within or outside the foot is observed. The present analysis departs from previous accounts of HVD that considered the role of prosodic structure in that it posits (a) that the process is conditioned by alternating rhythm (contra Cedergren Reference Cedergren and Sankoff1986 and Bosworth Reference Bosworth2017); (b) that two types of parsing are available for QF words (contra Verluyten Reference Verluyten1982, Bosworth Reference Bosworth2017 and Garcia et al. Reference Garcia, Goad, Guzzo, Jesney, O’Hara, Smith and Walker2017); and (c) that word size and word edges influence the process (contra Cedergren Reference Cedergren and Sankoff1986 and Verluyten Reference Verluyten1982).

Turning to future research, it should be noted that neither of our experiments included words with schwa or morphological complexity. Concerning the former, both high vowels and schwa are subject to deletion in QF, although the mechanisms regulating deletion may be different for each of them. Future research should explore how HVD interacts with the presence of word-internal schwa, and whether deletion targeting schwa has a similar prosodic conditioning. Concerning the latter, future research should also probe the extent to which morphological structure may impinge on the prosodic conditioning of HVD, given that many suffixes in French begin with the vowel .