2.1 Sonority-sensitive stress

Syllable weight factors into several core domains of phonology, including stress, tone, prosodic minimality, syllable structure, compensatory lengthening, meter/textsetting, end-weight, morphophonology, and others. For binary weight distinctions, two criteria prevail, with roughly equal frequency: on the first, a syllable is heavy iff it contains a long vowel (usually including diphthongs). That is, schematically, VV > {VC, V} (ignoring the onset). The other most frequent criterion treats a syllable as heavy if it has a long vowel or coda, that is {VV, VC} > V. Beyond these two criteria, in some systems, sonority affects weight, such that a more sonorous rime (e.g., one with a lower vowel, or a sonorant coda) counts as heavier.² That said, the prevalence of sonority sensitivity varies across weight-based phenomena. Sonority (at least sonorant vs. obstruent) can clearly condition weight for tone licensing (Gordon 2006: 32f), end-weight (Ryan 2019b: 169–77), meter (ibid.: 147–52), and syllable structure (Zec 1995: 101; Sayeed 2017). Less frequently, it conditions weight for compensatory lengthening (e.g., Kung 2007: 139f; Kiparsky 2011), reduplication (e.g., Ryan & Heath 2022), allomorphy (e.g., Anttila 1997: 5f), and perhaps minimality (cf. Ryan 2019b: 122f, 245f).

Stress, for its part, is among the weight-based phenomena for which sonority-sensitivity is less frequent. Though sonority-driven stress was once fairly uncontroversial (e.g., Hayes 1995; Kenstowicz 1997; de Lacy 2004; Gordon 2006) — Munshi & Crowhurst (2012: 429) could still refer to it in an “impressive array of languages” — more recently, several claimed cases have been reanalyzed (e.g., Rasin 2018: 44–7 on Kobon) or refuted empirically (e.g., Shih 2018 and Bowers 2019 on Gujarati), leading several scholars to call its existence in general into question. For example, Rasin (2018: 13) hypothesizes: “The distribution of stress is never conditioned by segmental features,” including [±sonorant]. Likewise, Shih & de Lacy (2019: 11) (also Shih 2016) conclude: “there is no reliable evidence that metrical feet are attracted to or repelled by segments of particular sonority levels” (but see the next paragraph for qualifications). Faust & Ulfsbjorninn (2025: 19–21) too propose a reanalysis of a case in which sonorants outweigh obstruents for stress (namely, Kwak’wala; see below) and conclude that the existence of sonority-driven stress in the general case is tenuous. But not all scholars agree: Paramore (2025), for instance, allows stress to be sensitive to the sonorant vs. obstruent distinction even when both are moraic (e.g., V_μN_μ > V_μT_μ; similarly Ryan 2019b: 19).

Accounts that deny sonority-driven stress nevertheless qualify that claim in several ways. First, none treats weight distinctions between reduced and full vowels as sonority-based. Second, some of the accounts explicitly allow sonority to influence stress indirectly. For example, Rasin (2018: 21, 83) observes that when Latin stress differs as a function of consonant sonority (e.g., Latin [woˈlup.taːs] ‘pleasure’ vs. [ˈwo.lu.kris] ‘bird’), it is because sonority conditions syllabification (which in turn conditions stress), not because stress is directly sensitive to sonority.³ Third, Faust & Ulfsbjorninn (2025: 21) make explicit that they are concerned only with categorical rules, and the same is implicit in other accounts, none of which engage with experimental evidence for sonority sensitivity (e.g., Carpenter 2010). Fourth, Rasin (2018) and Shih & de Lacy (2019) explicitly confine their claims to stress, setting aside other weight-sensitive phenomena. Fifth, Rasin (2018) and Shih & de Lacy (2019) focus almost exclusively on vowel sonority. Thus, some cases in which stress appears to be determined by consonant sonority (next paragraph) go unaddressed.⁴ Moreover, even for vowel sonority, perhaps not every case has been reanalyzed or refuted empirically (e.g., Kiriwina/Kilivila and Mordvin from de Lacy 2004, Cowichan from Bianco 1994). Finally, although some classic examples of sonority-driven stress are claimed to invoke gradient sonority scales (e.g., low > mid > high; Kenstowicz 1997), the literature also treats binary distinctions in sonority (e.g., an alleged low vs. non-low criterion in Gujarati) as cases of sonority-driven stress. This last point is relevant to the present article, which argues that Portuguese stress is sensitive to a binary distinction in sonority.

Since interactions between stress and coda sonority are the focus of this article, this paragraph addresses their typology. Several potential cases appear in grammars (e.g., 3 of 107 languages surveyed by Gordon 2006), though some are underdocumented, and some potentially dubious. For example, per Levinsohn (1976: 29), Inga Quechua stress falls on the ultima only if it has a sonorant (not obstruent) coda. Nevertheless, the evidence is lacking: few examples are provided; closed ultimas are in general uncommon; and we are not aware of any corroborating phenomena or sources. Some other reported cases are in need of further research or of uncertain relevance here, including Lamang (Wolff 1983: 19), Orya (Fields 1991: 32f), Paipai (Joel 1966: 10), and Yahi (Hinton & Luthin 2002). Better established cases include Totonacan languages such as Huehuetla Tepehua (Kung 2007) and Pisaflores Tepehua (MacKay & Trechsel 2013). In these languages, primary stress is penultimate unless the ultima has a sonorant or laryngeal (but not obstruent) coda. Kraehenmann (2003: 170ff) likewise maintains that syllables ending with sonorant (but not obstruent) consonants attract stress in Thurgovian Swiss German.

Wakashan languages furnish perhaps the best-known case of coda sonority-driven stress (being, for instance, the only such case discussed by Paramore 2025 and Faust & Ulfsbjorninn 2025), though it too is debated.⁵ In Kwak’wala, stress falls on the leftmost heavy syllable (or, if there is no heavy, the ultima), where a heavy contains a full vowel or a non-glottalized sonorant coda (Zec 1988; 1995; Bach et al. 2005). Elfner (2023) argues that the case reflects phrase-level prominence rather than metrical stress, but either way, “prominence placement is driven by sonority” (74). Faust & Ulfsbjorninn (2025: 20), however, propose a reanalysis that obviates reference to sonority: as they note (with Noguchi 2011: 323–5), a plain sonorant coda—the only type of “heavy” coda—always follows a schwa. This schwa, they maintain, can be analyzed as epenthetic/phonetic, such that the sonorant consonant is the nucleus. Stress then targets “the leftmost nucleus with lexical features” (i.e., full vowel or syllabic consonant), without reference to weight. The case remains open, however, as the reanalysis faces some potential issues. First, the reanalysis might not be available for related Makah or Nuu-chah-nulth, which have the same coda obstruent vs. sonorant criterion, but lack reduced vowels (Wilson 1986: 288f). Second, as Elfner (2023: 18f, 26) makes explicit, Kwak’wala permits Vʔ rimes (where V is a full vowel) and they count as light for stress (e.g., [hə.m’aʔ.ˈd^zu] ‘long soft objects’); on Faust & Ulfsbjorninn’s (2025) proposal, Vʔ is erroneously predicted to bear stress like any other full-voweled syllable. Finally, the reanalysis (as given) does not explain why glottalized sonorant codas, which must (like plain sonorant codas) be preceded by schwa, do not likewise pattern as nuclei for stress assignment.

In summary, while different scholars operationalize “sonority-driven stress” in different ways, its existence in general is controversial. Moreover, most of the literature on the subject has focused on vowel sonority. Recent articles such as Faust & Ulfsbjorninn (2025) and Paramore (2025), however, have brought the question of whether stress can be conditioned by coda sonority back to the theoretical forefront. Reported cases in grammars are either underdocumented (e.g., Inga Quechua) or have been challenged (e.g., Kwak’wala). Against this backdrop, we turn to BP, adding new evidence about the productive application of stress to nonce words with various weight patterns and demonstrating that at least in the ultima, coda sonority determines stress placement. This is the first study to address coda sonority-driven stress experimentally, complementing the (relatively sparse) typological evidence by demonstrating its productivity in Portuguese.

2.2 Stress in Portuguese

The stress system in BP is virtually identical to that of European Portuguese. We therefore make no distinction between the two varieties, even though our data comes exclusively from the former. In this language, stress normally falls on the ultima when it is heavy (papél ‘paper’). Otherwise, stress falls on the penultimate syllable (cabíde ‘hanger’) — this is the so-called regular pattern, which accounts for approximately 70% of non-verbs in the language (Bisol 1992; Araújo 2007; Wetzels 2007; Garcia 2014). Antepenultimate stress is also attested (e.g., abóbora ‘pumpkin’), and has traditionally been treated as exceptional in non-verbs, though studies have identified systematic phonological and morphological generalizations in this domain (see Araújo et al. 2007; Hermans & Wetzels 2012; Garcia 2017). As in English, stress behaves differently in verbs, where it is heavily influenced by morphology (see Garcia 2025b for a recent review). Thus, it is non-verbs that provide the relevant empirical domain for testing whether coda quality affects stress preferences beyond the basic heavy/light distinction.

A heavy syllable in Portuguese contains either a diphthong or a coda consonant (vowel length is not phonemic in the language). The inventory of possible codas in the language is {s, N, l, r} (both word-internal and word-final), where N represents one of the following allophonic nasal consonants: [m, n, ŋ]. This inventory makes BP a useful test case because it provides a clear sonorant–obstruent contrast among codas, albeit one in which the obstruent side of the contrast is represented only by /s/. We show that coda /s/ behaves differently (at least in the ultima), i.e., it is less stress-attracting than the sonorant codas attested in the language. One member of the sonorant set, /l/, requires separate mention as it is typically semi-vocalized: a word like papél /papɛl/ is pronounced as [pa.ˈpɛw] by the vast majority of BP speakers.

A separate confound involving final /s/ is morphological: plural in Portuguese is represented by the morpheme /-s/. Since plural /-s/ does not affect the location of stress, the task of comparing obstruent and sonorant codas becomes less straightforward. Fortunately, there are monomorphemic words ending in /s/ in the lexicon, which can help us map potential sonority effects in the language as a whole. We return to this question after presenting and discussing the results in §4.

3.1 Participants

Participants (n = 28) were native speakers of BP living in Canada at the time of the experiment (18 females, 10 males). They shared relatively similar characteristics with regards to education, region of origin, and linguistic background. Most participants had a bachelor’s degree or a graduate degree (n = 23), and most came from the same two regions in Brazil, namely south and southeast (n = 22). The average age was 35 (s = 9). Almost all participants reported at least an intermediate proficiency level in English (20 reported advanced/fluent English). Some also reported some degree of fluency in Spanish and French, ranging from intermediate to advanced levels. These extralinguistic variables had no substantial impact on the patterns we discuss in this paper.

3.2 Design and materials

We designed two tasks using the Gorilla platform (Anwyl-Irvine et al. 2020) where we varied the quality of the possible coda consonants in the language: {s, N, l, r}. Task 1 was an auditory forced-choice task with different weight profiles (HLL, LHL, LLH, LLL). In this task, participants saw the orthographic form of nonce words (n = 86) and heard two versions of each word, i.e., minimal pairs where the only difference was the location of stress in the trisyllabic stress window. All items were recorded by a native speaker of BP with phonetic training. For LLH words, the contrast was always between final and penultimate stress, whereas for other weight profiles, the contrast was always between penultimate and antepenultimate stress. This allowed us to use penultimate stress as the reference level (versus alternative stress given weight profile) when modeling the data. Participants were asked to choose which auditory version of each nonce word they preferred, e.g., nulquibe (HLL) → [ˈnuw.ki.be] or [nuw.ˈki.be].

Task 2 was an orthographic task, where participants were also presented with nonce words (n = 50) displaying different weight profiles (HLL, LHL, LLL). Task 2 was included to test whether the same stress preferences would generalize across presentation modality, i.e., whether the sonority-sensitive pattern found in auditory stimuli would also emerge when participants inferred stress from orthographic forms alone. A carrier sentence was presented for every item (in Portuguese; translated here): The XXX will be released at the end of the month. Participants were always given a binary choice to fill in the blank. For example, “DIpante” versus “diPANte” (LHL). Like Task 1, Task 2 also focused on minimal pairs where the location of stress was the only difference. Participants were told that capital letters indicated “the strongest syllable in the word.”

3.3 Procedure

To ensure the reliability of online results, some participants participated in person (n = 11) in a sound-attenuated room at Université Laval. The remaining participants completed the experiment remotely and resided in different cities in Canada. No meaningful differences were found between these two groups, so our discussion below will not distinguish the two methods of participation.

In both Tasks 1 and 2, participants were told that the words being evaluated had no correct pronunciation, given that they do not exist in the language. In addition, they were told that all words were in singular form and represented an object. This is important to minimize readings where participants assume they are evaluating a verb or a plural noun in the case of items ending in /s/. The distribution of items by task, weight profile, and syllable shape is summarized in Appendix A, where we also provide the full list of stimuli; the corresponding materials are available in the OSF repository listed at the end of this paper. Finally, the experiment including both tasks took approximately 20 minutes to complete, not including a short questionnaire with demographic information completed before Tasks 1 and 2.

4.1 Morphology and final /s/

A potential concern is that the final-/s/ effect could be morphology-driven rather than sonority-driven. Portuguese plural is marked by /-s/, and plural /-s/ is invisible to stress: adding it does not shift stress or make the ultima stress-attracting. This creates a natural confound for comparing obstruent (i.e., /s/) and sonorant codas. We begin by noting that the experimental design disfavors plural readings, both in the instructions and in the use of carrier phrases incompatible with plural readings, but even so, the possible confound cannot be eliminated by design alone. This section therefore presents additional evidence that the observed sonority effect is not an artifact of participants interpreting /s/-final forms as plural. We begin with lexical evidence, followed by a discussion of individual participant behavior that applies increasingly stringent robustness checks.

First, it is important to note that monomorphemic /s/-ending words do exist in the Portuguese lexicon. Using the Portuguese Stress Lexicon as our baseline (Garcia 2014), we can easily map the stress patterns in the lexicon as a function of (final) coda quality. We can then overlay these patterns (monomorphemic words) with the patterns in the experiment — this is shown in Figure 5. Notice how the inversion observed in the experimental data is also observed in the lexicon. Simply put, monomorphemic words ending in /s/ are less likely to have final stress than words ending in sonorants in the lexicon. This is exactly what we observe in the experimental data discussed above.

Figure 5: Lexical patterns (semitransparent thick lines) and experimental results (dashed lines). No LHL word with APU stress and an obstruent coda is found in the lexicon, hence the absence of an orange line for LHL words.

Second, we can inspect the behavior of individual participants, shown in Figure 6. If some participants assume that coda /s/ is plural /-s/, their preferences should be virtually categorical. While there are indeed some participants (n = 8) who never choose final stress in /s/-ending words, the vast majority of participants display a non-categorical pattern of behavior. Crucially, none of the participants show the opposite effect to what we find for the data as a whole.

Figure 6: Individual preference for final stress as a function of coda sonority: obstruent coda in orange, sonorants in violet.

As a robustness check, we reran the sonority model excluding the eight participants who never chose final stress for /s/-final LLH items, i.e., the participants whose behavior is most compatible with a categorical plural-reading account. The critical LLH contrast remained credible: /s/-final items continued to favor penultimate stress more than sonorant-final items ( $\widehat{\beta }=2.30$ , 95% CrI = [1.46, 3.13]). The corresponding HLL and LHL contrasts remained non-credible (HLL: $\widehat{\beta }=-0.56$ , 95% CrI = [−1.32, 0.19]; LHL: $\widehat{\beta }=-0.63$ , 95% CrI = [−1.42, 0.17]). Thus, the main sonority effect does not depend on the participants whose behavior is most plausibly attributable to a morphological effect.

Excluding participants, however, treats plural-reading as all-or-nothing, whereas Figure 6 suggests that it is a matter of degree: a given participant might read only a subset of /s/-forms as plurals, in which case the confound would still apply to the non-categorical majority. To model this directly, we fit a hierarchical two-component mixture with participant-specific mixing weights. The model estimates, for each participant j, a latent probability π_j of their interpreting /s/-forms as plurals. Under this model, the probability of a penultimate (PU) response on an /s/-final LLH trial is π_j + (1 − π_j)θ_i, where θ_i carries the usual regression structure (e.g., sonority effect, item-level variability) for the purely phonological grammar (for non-/s/-forms, only θ_i applies). The critical question is whether the obstruent-sonorant contrast survives in θ_i once π_j has absorbed whatever morphological inflation it can.

It does. The model estimates non-trivial plural-reading: the population-mean π is 0.38 (95% CrI = [0.10, 0.72]), with per-participant π_j varying continuously across [0, 1], in line with Figure 6. The singular-reading (i.e., pure phonology) component, however, still shows a large sonority effect in LLH: /s/-final items are predicted to receive PU on roughly 65% of trials when no plural-reading is happening, versus only 23% for sonorant-final items ( $\widehat{\beta }=1.91$ , 95% CrI = [0.74, 2.93]). The observed rates are 79% and 26%, so morphology accounts for less than a quarter of the asymmetry, and most of the gap remains once it is partialed out (see Figure 7). Because the mixture attributes as much variance to morphology as its structure allows, the residual obstruent-sonorant gap in θ is a lower bound on the contribution of phonology.

Figure 7: Mixture-model decomposition of predicted P(penultimate) on LLH trials, by ultima coda: phonological component θ (dark) plus plural-reading lift π · (1 − θ) (light, only applies to /s/). Error bars are 95% CrIs on θ; orange ticks are raw observed proportions (slightly offset from bar tops, which are fixed-effects predictions).

5.1 Morphology

Taken together, the evidence above limits what a morphology-based account can explain. The lexical comparison shows that the same asymmetry is present among monomorphemic words, where final /s/ is not a plural exponent. The participant-level pattern shows that only a minority behave in the categorical way expected if final /s/ is parsed as plural /-s/. A model excluding those participants still yields the same critical LLH contrast, as does a model including all participants, but allowing them to vary in their propensities to parse /s/-forms as plural. This does not eliminate morphology as a possible factor for a subset of responses, but it does indicate that the main effect cannot be reduced to plural marking. We also cannot eliminate the possibility that morphology might play an indirect role. One could imagine, for instance, that a preponderance of plural forms in the learning data (alongside /s/-final monomorphemes) might help learners to generalize a rule to the effect that final /s/ is non-moraic. In this manner, the phonological treatment of /s/ might not be wholly independent of the morphological status of /s/ (though, as established, even monomorphemes ending with /s/ are much less likely than those ending with sonorants to take final stress).⁶ This question concerns the source of the generalization — that is, diachrony or learnability — rather than its synchronic status, and thus is beyond the scope of the present experimental study.

5.2 The phonological representation of final /s/

A different concern is that BP final /s/ may be structurally special independently of morphology. On this view, the contrast would not be between sonorant vs. obstruent codas as such, but between ordinary codas and word-final /s/ with a distinct (e.g., appendix) structure. BP cannot fully resolve this question, since the language has only one obstruent coda. We argue in this section, however, that the appendix analysis is not in conflict with the sonority account. A fleshed-out version of the appendix analysis must answer why /s/, rather than codas in general, is treated as an appendix, and the best-motivated answers, we argue, involve sonority. Thus, far from undermining sonority-based weight, appendixhood is one possible implementation of it.

Two factors make the appendix analysis natural to consider for word-final /s/ in BP. First, of all segments, /s/ is cross-linguistically the most likely to be analyzed as an appendix, although most cases involve /sC/ onsets or /Cs/ codas (Kaye 1992; Vaux & Wolfe 2009; Goad 2011). Second, appendixhood is typologically concentrated at word peripheries (Vaux & Wolfe 2009). The relevant “light” segment in BP — word-final /s/ — combines both these properties. Even though no study has, to our knowledge, proposed that BP final /s/ is better described as an appendix, we do not rule out that possibility. However, we argue below that, like other cases of /s/ appendices, sonority still plays a key role under that analysis.

We propose that word-final /s/ (realized as [s] or [ʃ]) is non-moraic in BP. Nevins (2015: 310–2) proposes the same based on independent evidence: aside from the comparatively low rate of final stress in /s/-final monomorphemes already noted, he observes that the dialect of Rio de Janeiro epenthesizes [j] before final [ʃ] iff stress falls on the ultima (compare [a.veʃ.ˈtɾujʃ] ‘ostrich’ to [ˈbo.nuʃ] ‘bonus’), and attributes this asymmetry to Stress-to-Weight (a stressed syllable must contain two moras). In this case, he argues, [j]-epenthesis supplies the second mora, satisfying Stress-to-Weight, but the constraint has no effect on unstressed ultimas. This analysis depends on final /s/ being non-moraic, which Nevins (2015) posits as a general property of BP: “sibilants are not sonorous enough to be moraic.” Our conclusion is essentially the same, though it hinges on independent arguments.

For present purposes, we will assume that all word-internal codas, including /s/, are moraic. The present experimental results fail to distinguish /s/ from other codas in non-final syllables, and weight sensitivity in non-final syllables is documented elsewhere (e.g., Garcia 2019). Moreover, the localization of the present sonority effect to word-final position is itself expected: BP stress is primarily weight-sensitive in the ultima (§2.2), so any sonority-driven contribution to weight should manifest most clearly there. The treatment of non-final codas is not, however, critical to what follows.

Non-moraic word-final /s/ is compatible with various structural representations. As schematized by Figure 8, the adjunction site of /s/ might be the mora (μ), syllable (σ), or prosodic word (ω).⁷ In any case, the resulting syllable is monomoraic. Both (a) and (b) are widely employed by phonologists as the unmarked representations of monomoraic CVC syllables (e.g., (a) by Hyman 1985; Hayes 1989; 1995; (b) by Rosenthall & van der Hulst 1999; Morén 2001). Rosenthall & van der Hulst (1999: 503) refer to the coda in (b) as an appendix, which they take to violate the constraint *Append. As they further note (following Sherer 1994), *Append can be regarded as a family of constraints penalizing each of (a–c): *Append(to-μ), *Append(to-σ), and *Append(to-ω), respectively. Note that in (b), /s/ is part of the syllable, and hence a coda, but may also be regarded as an appendix, as Rosenthall & van der Hulst (1999) do.

Figure 8: Possible adjunction sites of /s/.

A word-final consonant that is ignored for weight is traditionally described as extrametrical (or extraprosodic) and notated by angled brackets: CV<s>. In Optimality Theory, extrametricality is usually not taken to have a formal status beyond the sorts of adjunction structures in Figure 8 (that is, angled brackets are not formal parts of representations). For example, Kager (1999: 268) implements final consonant extrametricality using *Final-C-μ, which penalizes a moraic final consonant; both (b) and (c) (and possibly (a)) satisfy this constraint.

Given this background, one possible analysis of word-final /s/ takes it to be an appendix to the prosodic word, and hence non-moraic, whereas all other codas — whether word-final sonorants or word-internal — are parsed into the syllable, and hence moraic. In the formalism of Rosenthall & van der Hulst (1999), *Append(to-σ) is top-ranked: any coda that is parsed into a syllable must be moraic, as illustrated by tableau (1). (The constraint Weight-by-Position would have the same effect.)

For word-final consonants, however, an additional way to satisfy *Append(to-σ) becomes available: the consonant may be an appendix to the prosodic word, as shown for /s/ in tableau (2). Because ω-adjunction is marked, some constraint must dominate *Append(to-ω) to compel it. Here, we employ *T_μ, a constraint penalizing moraic obstruents (Zec 1988; 1995; Nevins 2015). Rosenthall & van der Hulst (1999: 502) likewise explicitly acknowledge this constraint.⁸

As desired, a word-final sonorant, /n/ in tableau (3), remains moraic. Because *T_μ does not penalize moraic [n], nothing prevents it from being parsed into the syllable.⁹

While not the only possible analysis, this analysis illustrates that the possibility of ω-appendixhood does not in itself oppose the sonority-based explanation. On the contrary, sonority can explain why /s/, on such an analysis, is an appendix. That said, if one were to substitute a hypothetical constraint such as *SibCoda (penalize a sibilant coda) in place of *T_μ above, the analysis would still work. *SibCoda, like *T_μ, would have the effect of pushing the sibilant out of the syllable. However, *SibCoda is more ad hoc: it singles out a phoneme (ironically, the only obstruent that is allowed as a coda in BP), has less typological backing than *T_μ, and lacks independent support, in the sense that no one to our knowledge has proposed that word-final /s/ is not a coda in BP, as mentioned earlier. If *X-Coda constraints can be freely stipulated, why do we not find languages in which, say, *SonCoda (or *n-Coda) drives sonorants (or just [n]) into appendix position? In the typology of weight-sensitive phenomena (§2.1), languages restrict coda weight along the obstruent/sonorant divide, not along the sibilant/non-sibilant or any other arbitrary divide. Moreover, previous discussions of /s/-extrametricality in BP (e.g., Kaye 1992; Goad 2011) concern only /sC/ onset clusters, where appendixhood is motivated by the Sonority Sequencing Principle.¹⁰

Still other analyses are conceivable. For example, Paramore (2025: 16) argues that codas are universally moraic, and accounts for coda sonority-driven stress using the constraint S→[μ_Rμ_R]_σ, which requires stress (S) to fall on a syllable with two sonorant moras (μ_R). With such a constraint, all codas could be analyzed as moraic in all positions in BP (i.e., simple weight-by-position). This approach would need to be fleshed out for BP in order to limit the distinction between codas to the ultima. At any rate, this line of analysis once again adopts a sonority-driven stress approach.

Taken together with the morphological argumentation in §4.1, we take the BP data to support a limited but robust conclusion: final sonorant codas are more stress-attracting than final /s/ in BP nonce words, along with the lexicon, and this asymmetry resists reduction either to plural morphology or to a sonority-free appendix account. Indeed, the localization of the asymmetry to final position — the same position where weight is otherwise most relevant to BP stress — is what a sonority-driven weight account predicts. While word-peripheral /s/ is analyzed as an appendix in some languages, its appendixhood in well-studied cases is not merely stipulated (“sibilants prefer to be appendices”), but independently motivated by sonority,¹¹ and we argue that the same applies here if an appendix analysis is employed.

5.3 Brazilian Portuguese and sonority-driven stress

Finally, we return to the question of how the present case relates to the recent skepticism about sonority-driven stress. The BP pattern — stress placement is conditioned by whether the final coda is a sonorant or an obstruent — is sonority-driven stress in the broad descriptive sense, and it can be analyzed as such directly (e.g., using Paramore’s 2025 constraint S→[μ_Rμ_R]_σ, as discussed in §5.2). Nevertheless, in §5.2, we also considered an appendix-based analysis, in which *T_μ causes word-final /s/ to be realized as a ω-appendix, and thus non-moraic. Whether this latter analysis fits the definition of “sonority-driven stress” depends on how the analyst defines the term.

Recall from §2.1 that Rasin (2018) proposes the stress encapsulation universal, according to which “[t]he distribution of stress is never conditioned by segmental features” (including sonority). Nevertheless, Rasin (2018) allows sonority to influence stress indirectly, via syllable structure, and does not count such cases as violating the proposed universal. As such a case, he gives the example of Latin [woˈlup.taːs] ‘pleasure’ vs. [ˈwo.lu.kris] ‘bird’, where sonority determines how the interlude ([p.t] vs. [.kr]) is syllabified, and hence the weight of the penult. While one might expect that the appendix analysis of BP would be regarded as another case of the same indirect conditioning — sonority conditions prosodic structure, which in turn conditions stress, in parallel with the Latin case — in practice, Rasin (2018: 20–2) employs a CV tier rather than arboreal prosodic phonology, and does not address phonological appendices,¹² so we will not speak for how the appendix analysis in §5.2 might map onto that approach. At any rate, we acknowledge that the appendix analysis of BP is at least consistent with the spirit of the argument that structural considerations can obviate direct reference by stress to sonority.

Similarly, Shih & de Lacy (2019), in arguing against the existence of sonority-driven stress, circumscribe the phenomenon narrowly (see §2.1), allowing that “sonority can indirectly affect metrical structure: if segments of certain sonority are required to have particular moraic configurations and metrical structure is appropriately sensitive to moraic content, then segmental sonority can appear to influence metrical structure” (10). This distinction between stress being mediated by sonority vs. moraicity is especially meaningful for vowel-sonority cases, which are the focus of Shih & de Lacy (2019). If all vowels are moraic, but lower vowels nevertheless pattern as more stress-attracting — or if there is a gradient effect such as low > mid > high — then constraints on stress/feet must directly invoke sonority.

In the context of coda sonority, however, the same operationalization of direct vs. indirect is less probative, since consonants can more readily be analyzed as differing in moraicity. In particular, any VN > VT criterion for stress that can be analyzed with N alone being moraic (e.g., via *T_μ) does not count as “sonority-driven stress” under this definition. Qualifying cases would have to include both a VN > VT criterion for stress and independent evidence that the stress system treats both N and T codas as moraic. While such cases are conceivable and have even been put forth,¹³ this is a high bar, and yields a narrower sense of “sonority-driven stress” than the prevailing usage in the context of consonants. For example, Kwak’wala (§5.2), a canonical putative case (cf. e.g., Rasin 2018: 83, Faust & Ulfsbjorninn 2025: 20), would not count as such, provided that sonorant codas are simply deemed moraic. This divergence between the narrow technical sense vs. common descriptive sense of “sonority-driven stress” matters for how claims about the typology are received: a reader who interprets “sonority-driven stress does not exist” in the broad sense risks concluding that VN > VT patterns are unattested, when analysts’ actual claim is the narrower one that such patterns are mediated by moraic structure.

In sum, the classification of BP depends on what one takes “sonority-driven stress” to require. On a broad descriptive use of the term, BP is plainly a case: stress preferences differ according to the sonority of the final coda. In a narrower sense, there are analyses available for BP in which sonority conditions moraicity, which in turn conditions stress, such that constraints on stress need not directly reference sonority. Setting aside questions of how exactly sonority affects stress placement in various frameworks, however, it bears repeating that the typology furnishes relatively few cases in which the sonority levels of consonants descriptively affect stress placement. There is even less experimental investigation of such cases. The present article contributes new evidence on both fronts, furnishing experimental evidence for a productive consonant-sonority asymmetry in BP.