Resyllabification turns a word-final coda into a derived onset in prevocalic contexts. For Spanish, Harris and Kaisse (1999, p. 137) formalize an autosegmental operation (1) that delinks a segment Y from the rhyme Nʹ and reassociates it as an onset to the following syllable Nʺ. Resyllabification is complete in that delinking and reassociation render derived and canonical (i.e. word-level) onsets structurally identical.

An argument often cited by generative phonologists in support of complete resyllabification comes from Hualde’s (1991) rule-based account of fricative lenition in the traditional Chinato Spanish (henceforth, CS) variety of western Spain. In the postlexical (i.e. phrase-level) phonology, an aspiration rule targeting coronal obstruents in the syllable rhyme is ordered to apply after resyllabification. Because resyllabification destroys its triggering environment, the aspiration rule applies transparently, affecting only canonical codas. For the analysis to work, derived onsets must be structurally identical to canonical onsets, with no remaining association to the preceding syllable rhyme. ¹ Even since the rise of Optimality Theory (henceforth, OT; McCarthy & Prince, 1999; Prince & Smolensky, 2004), which abandons rule ordering and serial derivations in favor of parallel evaluation of output candidates by surface-oriented constraints, most analyses of Spanish phonology still take the structural identity of derived and canonical onsets as a given. To our knowledge, no alternative account of the CS data has been proposed in the OT literature thus far.

More recently, however, the structural identity assumption has been problematized by experimental acoustic studies of Northern-Central Peninsular Spanish (henceforth, NCPS) that uncover subphonemic differences in the realization of intervocalic /s̺/, which is shorter and more voiced in word-final position than in word-medial or word-initial position (Hualde & Prieto, 2014; Strycharczuk & Kohlberger, 2016; Torreira & Ernestus, 2012). How to reconcile the phonological evidence in favor of complete resyllabification with the phonetic evidence against it remains a puzzle.

In this article, we first propose an articulatory representation of ambisyllabicity that predicts the shortening of word-final prevocalic /s̺/ as observed in NCPS. The representation is couched within a split-gesture, competitive, coupled oscillator model of the syllable in Articulatory Phonology (Burroni, 2022; Goldstein et al., 2006; Goldstein & Pouplier, 2014; Hoole & Pouplier, 2015; Marin & Pouplier, 2010; Nam, 2007a, 2007b, 2007c; Nam et al., 2009; Tilsen, 2017; Zsiga, 2011, 2020, pp. 202‒210). Then, we show how an ambisyllabic gestural coupling representation streamlines the amount of phonological computation required to account for coronal fricative lenition in CS as compared to alternative theoretical approaches. We offer an OT analysis that also predicts typological patterns of coronal fricative aspiration and voicing that are observed across contemporary Spanish varieties. Lastly, an examination of the consequences of ambisyllabicity for the analysis of rhotic consonants leads us to argue that a markedness constraint on the perceptibility of rhotic contrast is responsible for rhotic neutralization in Spanish phonology, which altogether lacks a rule or constraint targeting individual coda rhotics (Bradley, 2005b, 2006b, 2020).

This article is organized as follows. Section 2 gives the relevant historical background and data from CS and presents Hualde’s (1991) analysis motivating complete resyllabification. Section 3 reviews the phonetic evidence from NCPS that resyllabification is incomplete. Section 4 proposes an ambisyllabic, competitively coupled split-gesture representation of derived onsets in NCPS. Section 5 presents our OT account of coronal fricative lenition in CS and explores its typological predictions. Section 6 examines rhotic contrast and neutralization in NCPS. Section 7 concludes.

We begin with some background on the historical evolution of Spanish anterior coronal sibilants. In Medieval Spanish (henceforth, MS), the voiced dentoalveolar affricate /dz/ and apicoalveolar fricative /z̺/ were in phonological contrast with their voiceless counterparts /ts/ and /s̺/. ² Deaffrication of the dentoalveolars resulted in fricatives /z̪/ and /s̪/. According to most accounts, the voiced sibilants /z̪/ and /z̺/ underwent devoicing and merger with their voiceless counterparts, after which point /s̪/ was interdentalized, giving rise to the contrast between interdental /θ/ and apicoalveolar /s̺/ in present-day NCPS (Hualde, 2014, pp. 150‒154; Núñez-Méndez, 2021, pp. 29‒31, 35‒36, 49; Penny, 2014, pp. 120‒123; Rost Bagudanch, 2022). MacKenzie (2022) proposes a revised model of the genesis of /θ/, based on a quantitative analysis of corpus data tracking the change from orthographic <d> to <z> in word-medial preconsonantal contexts, e.g. juzgar for iudgar ‘to judge’ (< Latin iudicare). He argues that the deaffricated reflexes of MS /dz/ and /ts/ underwent dissibilation to non-sibilant /ð/ and /θ/, respectively, before the devoicing process was completed by the 17th century-a view also held by Menéndez Pidal (1987, p. 113). ³

In CS, which is the traditional variety of Malpartida de Plasencia in the province of Cáceres, in Extremadura, Spain (Catalán, 1954; Espinosa, 1935; Hualde, 1991), the anterior coronal sibilants of MS evolved differently compared to NCPS, giving rise to some unexpected correspondences:

These and other similar data come from “texts sent by a native speaker of the dialect, doña Gregoria Canelo, to Ramón Menéndez Pidal in 1904, and that were published and studied by Catalán (1954)” (Hualde, 1991, p. 57), as well as from descriptions by Espinosa (1935). We have adapted Canelo’s orthographic renditions into modern IPA. In CS, both [θ] (2)a,b and [ð] (2)c,d appear as canonical onsets in word-medial intervocalic position. Surprisingly, NCPS [s̺] corresponds to interdental [θ] (2)a and [ð] (2)c in CS, which altogether lacks apicoalveolar fricatives. According to Hualde’s (1991, pp. 59‒60) historical account of CS, MS apicoalveolar /z̺/ and /s̺/ merged with the deaffricated reflexes of dentoalveolar /dz/ and /ts/, which then were fronted to /ð/ and /θ/, respectively. However, /ð/ did not devoice. CS therefore maintains the MS voicing contrast but neutralizes the place contrast to interdental. The contrast between /ð/ and /θ/ lends further support to MacKenzie’s revised chronology of MS sibilants. In fact, MacKenzie (2022, p. 191, note 15) cites CS as an example of a variety in which dissibilation of /z̪/ and /s̪/ took place without subsequent devoicing, while at the same time highlighting the loss of apicoalveolar /z̺/ and /s̺/. Word-initially, [θ] and [ð] are not in direct contrast in CS. Of the two fricatives, only [θ] appears after a pause (2)e,g. Espinosa (1935, pp. 78, 102) gives two examples of voiced plosive [d], after a pause (2)f and a nasal (2)g. Apparently, the dissibilation of /z̪/ to /ð/ in CS led to a phonemic merger between /ð/ and /d/, whereby /ð/ acquired plosive allophones in the same phonological contexts as /d/ (Hualde, 1991, pp. 60−61).

In the synchronic CS grammar, the contrast between /ð~d/ and /θ/ is neutralized by processes of voicing and aspiration outside of the canonical onset environment. Spaces appear in the phrasal examples below to make the word divisions clearer. First, [ð] appears in word-final position before a following vowel (3)a,b and, according to Espinosa (1935, pp. 61, 73), in stem-final position before a vowel-initial plural (3)c or derivational (3)d suffix (cf. NCPS [rés̺es], [més̺es], and [luθéɾo].) Second, the glottal fricative [h] appears before a consonant, within the same word (3)e or across word boundaries (3)f, and before a pause (3)a,c, where it variably disappears in utterance-final position, e.g. [lað álah ~ lað ála], [réðeh ~ réðe], etc.

Hualde (1991) formalizes two rules targeting coronal obstruents in the syllable rhyme. Coronal Obstruent Voicing (COV) operates as a cyclic, stem-level lexical “constraint or filter on representations that checks the well-formedness of certain segments in certain positions, rather than as an active feature-changing rule” (p. 65). Applying vacuously to underlying final /d/ but actively to final /θ, t/ in loanwords, COV explains why the three-way coronal contrast in [éθo] ‘that’ (2)a, [kóða] ‘thing’ (2)b, and, e.g. [pandeɾéta] ‘tambourine’, is neutralized to intervocalic [ð] in word- and stem-final environments (3)a‒d. The second rule is Coronal Obstruent Aspiration (COA). It applies postlexically to delink supralaryngeal features in the syllable rhyme, leaving behind only the laryngeal features responsible for aspiration, which is transcribed as the glottal fricative [h].

The sample derivations in (4) and (5) illustrate the analysis. Periods denote syllable boundaries, and the + symbol, the morphological boundary between a stem and a derivational affix.

For [lað álah] (3)a, syllabification first takes place within each word separately (4)a, which puts each /d/ in the syllable rhyme. Checkmarks indicate the vacuous application of COV (4)b. Postlexical resyllabification turns the word-final prevocalic coda into a derived onset (4)c. Ordered after the spirantization of postvocalic voiced plosives (4)d and the devoicing of the word-final coda obstruent (4)e, COA applies (4)f, followed by an optional rule of utterance-final glottal fricative deletion (not shown here). The derived onset fricative escapes COA, which applies only to the canonical coda fricative. For [luðéɾo] (3)d, the first cycle of the lexical phonology induces syllabification within the domain of the stem (5)a, where final /d/ in the syllable rhyme satisfies COV (5)b. Adding the derivational suffix triggers a second cycle: stem-final prevocalic /d/ syllabifies as a word-medial onset (5)c, and then COV fails to apply (5)d. The only postlexical rule that can apply in this derivation is Spirantization (5)f. The stem-final canonical onset fricative escapes COA, as does the word-final derived onset fricative (4)f.

COV and COA interact with resyllabification in opposite ways. First, COV is phonologically opaque because the rule ‘overapplies’ in onsets. Overapplication opacity arises because COV is ordered to apply before stem-final codas become derived onsets postlexically (4)c or canonical onsets on a second lexical cycle (5)c. This type of relationship, known as a counterbleeding rule order, is seen as evidence supporting the existence of derivational steps and abstract intermediate representations. Overapplication opacity is problematic for ‘classic’ monostratal OT, in which input-output mappings are evaluated in parallel without serial rule application or intermediate derivational steps (Baković, 2011, 2013). Second, because it is ordered after postlexical resyllabification, COA operates transparently, targeting only canonical codas (4)f. Applying first, resyllabification (4)c destroys the environment that triggers COA, thereby blocking the aspiration of derived onsets. Because COA could still apply if the word-final fricative were to maintain its lexical affiliation to the preceding rhyme, resyllabification must be assumed to be complete, i.e. derived onsets must be structurally identical to canonical onsets.

Contemporary laboratory phonology has come to question the generative assumption that resyllabification across word boundaries in connected speech is complete. The critique was initiated by phonetic studies of French enchaînement ‘linking’. Gaskell et al. (2002) report that French listeners show sensitivity to subphonemic differences between /VC#V/ and /V#CV/ sequences in lexical priming tasks (# denotes a morphological word boundary). According to Fougeron et al. (2003) and Fougeron (2007), French intervocalic consonants in laboratory speech are acoustically shorter as derived onsets than as canonical onsets. Listeners are capable of noticing durational and spectral cues in order to perceive the contrast between such sequences, which is incompletely neutralized (Fougeron, 2007). In American English laboratory speech, /l/ and /w/ are found to have less consonantal constriction and more asynchronous timing of intra-segmental tongue tip and tongue body gestures as derived onsets than as canonical onsets, e.g. hall otter vs. ha lotter, how otter vs. ha wadder (Gick, 2004). In British English, patterns of linguopalatal contact in derived onset /l/ differ from those found in canonical onset or coda /l/ (Scobbie & Pouplier, 2010). In American English spontaneous speech, intervocalic consonants are acoustically shorter word-finally than word-initially, e.g. beef eater vs. bee feeder (Tao et al., 2018).

Evidence for incomplete resyllabification in NCPS comes from acoustic studies reporting differences in the realization of apicoalveolar /s̺/ between vowels, depending on the presence and location of a word boundary. The following examples, from Strycharczuk and Kohlberger (2016, p. 5), illustrate the three main contexts of interest, namely canonical onset /s̺/ in initial (6)a and medial (6)b positions and derived onset /s̺/ in final position (6)c:

Torreira and Ernestus (2012) collected 1,257 tokens of intervocalic /s̺/, produced in spontaneous conversational speech by 27 female and 25 male speakers from Madrid as part of the Nijmegen Corpus of Casual Spanish, and analyzed voicing based on acoustic measurements of an uninterrupted pitch track during the fricative, as well as intensity differences with surrounding vowels. Of the total tokens of intervocalic /s̺/, 34% were realized with a fully voiced [z̺], more frequently in word-final position, the faster the rate of speech, and the shorter the duration of the fricative. Intervocalic /s̺/ voicing is thus consistent with laryngeal coarticulation, which is known to be sensitive to temporal variation. However, the fact “that voicing is considerably more common in word-final position suggests that this coarticulatory pattern is not an entirely passive phenomenon, and that it may contain the seeds of phonologization” (p. 139).

Hualde and Prieto (2014) analyzed 684 tokens produced by 16 female speakers from Madrid in directed conversational speech collected via map tasks for the Interactive Atlas of Spanish Intonation. Of the total tokens of intervocalic /s̺/, only 8.3% were fully voiced. Measurements of uninterrupted voicing replicated the findings of Torreira and Ernestus (2012) by revealing higher rates of fully voiced /s̺/ in final (12.5%) than initial (10.64%) or medial (5.93%) position, as well as significantly more voiced frames in final than medial /s̺/. Duration measurements revealed a significantly shorter intervocalic /s̺/ word-finally than initially or medially but no significant difference between the latter two contexts. Derived and canonical /s̺/ should be expected to pattern the same phonetically if Spanish resyllabification were completely neutralizing. Hualde and Prieto suggest that resyllabification is only partial and that a symbolic transcription of más amor ‘more love’ as [má.s̺a.móɾ] “simply offers an incomplete view of phonetic reality” (p. 124). They propose that in /VCV/ and /V#CV/ contexts, the onset consonant is “timed to start in-phase with the following vowel (Nam et al., 2009), whereas in /VC#V/ lexical syllabic affiliation may result in a different gestural coordination, even if, as it is generally assumed for Spanish and Catalan, there is postlexical resyllabification” (p. 123).

Evidence from a third study with carefully controlled materials uncovers even more systematic variation in laboratory speech. Strycharczuk and Kohlberger (2016) used a reading task with two-word stimuli embedded in carrier phrases, eliciting a total of 792 tokens from 1 male and 10 female speakers from northern and central Spain. As summarized in Table 1, an analysis of fricative duration revealed significant contextual differences, except for canonical coda /s̺/:

Between vowels, the fricative was the shortest as a word-final derived onset, but unlike in Hualde and Prieto’s results for spontaneous speech, there was also a difference between canonical onsets: initial /s̺/ was significantly longer than medial /s̺/. Canonical codas were shorter than canonical onsets, with no significant difference between preconsonantal contexts, e.g. meses pasados ‘past months’, seis españoles ‘six Spaniards’. No spectral differences are reported for the surrounding vowels, but a significantly longer preceding vowel in /Vs̺#V/ than in /Vs̺#PV/ (68.26 ms > 60.76 ms) is consistent with the cross-linguistic tendency toward open syllable lengthening, suggesting some type of affiliation between /s̺/ and the following vowel. The duration of derived onset /s̺/, intermediate between canonical onsets and codas, shows that resyllabification was nonetheless incomplete. The implication is that “structural differences between canonical and derived onsets are accessible to learners at a level of linguistic representation that connects directly to phonetic realization” (Strycharczuk & Kohlberger, 2016, p. 17).

Strycharczuk and Kohlberger consider two possible prosodic representations of incomplete resyllabification. If derived onsets are treated as ambisyllabic, then they should be distinguishable from canonical onsets and codas such that only ambisyllabic consonants will be targets of weakening. However, as the researchers point out, ambisyllabicity is also commonly used to represent intervocalic geminate consonants in many languages, which highlights the theoretical inconsistency of using the same type of syllabic structure to explain two opposing phenomena, i.e. lengthening and shortening. Jensen (2000) raises a similar theoretical critique of the use of ambisyllabicity in generative analyses of segmental strengthening and weakening in English, German, and Danish, as does Purse (2020, pp. 104‒105) based on an acoustic study of /l/-vocalization in Philadelphia English. As an alternative to ambisyllabicity for singling out derived onsets, Strycharczuk and Kohlberger suggest the possibility of prosodic misalignment, whereby /s̺/ associates to the following syllable onset while remaining associated to the preceding prosodic word. Prosodic word-initial /s̺/ is correctly predicted to be the longest because consonants are independently known to undergo domain-initial strengthening, which increases fricative duration (Fougeron & Keating, 1997; Keating et al., 2004; White et al., 2020)-an explanation we adopt here. However, as Strycharczuk and Kohlberger highlight, initial strengthening cannot explain why medial /s̺/ should be longer than final /s̺/, as neither fricative is word-initial. Although not statistically significant, in Table 1 the mean duration of coda /s̺/ is longer word-finally in /Vs̺#PV/ than word-medially /Vs̺PV/, which suggests the possibility of lengthening in prosodic word-final position. Unfortunately, final lengthening makes the opposite prediction that intervocalic /s̺/ should be longer word-finally than medially. How to capture the structural difference between canonical and derived onset /s̺/ in NCPS laboratory speech remains elusive.

Ramsammy (2021) used electropalatography to describe word-final /l/ reduction in the speech of four NCPS speakers from Galicia, based on data from a carrier phrase task involving nonce word stimuli. ⁴ Linguopalatal contact was the longest and most extensive utterance-finally before a pause but showed varying degrees of reduction before word-initial /a/ and voiceless plosives and fricatives. Because the study did not examine contexts of canonical onset /l/, it remains unclear whether participants showed asymmetrical reduction across /V#lV/, /VlV/ and /Vl#V/ contexts, as Strycharczuk and Kohlberger report for the duration of /s̺/. Still, Ramsammy argues that speaker-specific gestural overlap and reduction strategies provide a better explanation of the contextual variation in word-final /l/ than do differences in syllabification or prosodic structure, pace Rubach’s (1996) use of ambisyllabicity to explain properties of English word-final prevocalic /l/.

This section argues that the shortening of derived onset /s̺/ in NCPS, as documented by Strycharczuk and Kohlberger (2016), is predicted to emerge in speech planning from multiple competitive couplings among the fricative’s target and release gestures and the surrounding vowel gestures. In Articulatory Phonology (henceforth, AP; Browman & Goldstein, 1989, 1990), the grammar operates over dynamically defined movement tasks, or gestures, that instruct the active articulators to constrict the vocal tract during speech production. Abandoning the traditional phonological segment, AP posits that gestures are the ‘atomic’ units of speech production, glued together by coupling relations into larger ‘molecules’ that cohere into word forms during language acquisition (Goldstein et al., 2006, p. 225).

Building on work by Saltzman and Munhall (1989), the development of the coupled oscillator model of syllable structure (Goldstein et al., 2006; Goldstein & Pouplier, 2014; Hoole & Pouplier, 2015; Marin & Pouplier, 2010; Nam et al., 2009; Zsiga, 2011, 2020, pp. 202‒210) makes it possible to represent the temporal coordination of gestures using two basic timing relations. When two gestures G₁ and G₂ are coupled in-phase, they will begin simultaneously, but when coupled anti-phase, G₂ will begin after the midpoint of G₁. The model can be implemented computationally in the Task Dynamics Application (henceforth, TaDA; Browman et al., 2006; Goldstein et al., 2006). Coordination relations are formalized as intergestural coupling graphs, in which gestures are nodes that are connected by edges, or association lines, that specify the coupling relation. A solid line denotes in-phase coupling, i.e. G₁―G₂, while a dotted arrow denotes anti-phase coupling of the first gesture to the second one, i.e. G₁⇢G₂. Based on physical principles governing the movement of coupled oscillators, TaDA executes a speech planning process that computes the most stable timing pattern for all the gestures of an utterance at a specified speech rate. The output of the process is a gestural score, which displays gestures on separate tiers that correspond to the active articulators, i.e. Lips, Tongue Tip, Tongue Body, Tongue Root, Velum, and Glottis. Informed by mathematical equations of task dynamics, an interarticulator component then uses the gestural score to derive output movement trajectories that can serve as input for speech synthesis and vocal tract modeling (not examined here).

In doña Gregoria Canelo’s orthographic renditions of traditional CS speech (Catalán, 1954), <j> denotes a glottal fricative [h], and <d> denotes a voiced interdental continuant [ð] or its dental plosive allophone [d]. Canelo varies in her transcription of word-final intervocalic <d>, either attaching it to the following word, e.g. la dalaj for [lað álah] (3)a, or combining words without spaces, e.g. badamicada for [báð a mi káða] (3)b. An underappreciated observation about Canelo’s texts is that word-final fricatives are often transcribed as <j> followed by a space and then a word-initial <d>:

These examples correspond to NCPS casarnos este año, eres un confite and zarzas y las. While they may be transcription errors, Hualde (1991) acknowledges that such forms “could be interpreted as instances of ambisyllabicity: word-final [ð] links to the following syllable without delinking from the coda position that it occupied” (p. 64, footnote 7). Based on the orthographic variation in (12), we assume that CS derived onsets were ambisyllabic, as in (9). This assumption leads to a novel account of coronal fricative lenition, as we next show.

The patterning of word- and prefix-final prevocalic rhotic consonants in Spanish seems to provide further evidence that resyllabification must be complete. To indicate syllable boundaries, we add periods to the following examples, which are drawn from Hualde’s (2014, pp. 181‒185) description of NCPS:

There is a phonological contrast between an apicoalveolar tap /ɾ/ and trill /r/ in stem-medial intervocalic position, where both rhotics syllabify as single onsets, e.g. caro vs. carro (21)a. In the syllable rhyme, neutralization results in non-contrastive variation between the tap and trill, e.g. dar (21)b. Outside of the stem-medial context, the pronunciation of an intervocalic rhotic signals its morphological affiliation. For example, the trill is necessarily word-initial in the two-word phrase da rocas (21)c and stem-initial in the prefixed form extra-rápido (21)d, while the tap is necessarily word-final in dar ocas (21)e and prefix-final in super-ávido (21)f.

In a foundational rule-based account of Spanish rhotics, Harris (1983, 2001, 2002) argues that the stem-medial intervocalic contrast in (21)a is represented phonologically as a difference between a singleton /ɾ/ and its geminate counterpart /ɾɾ/. A lexical rule strengthens onset /ɾ/ to a trill [r] after a consonant in the preceding syllable rhyme (e.g. honra ‘honor’, alrededor ‘around’, Israel ‘Israel’) and is followed by a postlexical rule of coda /ɾ/-deletion before [r]. Together, these two rules derive an intervocalic singleton onset trill from the underlying stem-medial geminate /ɾɾ/, e.g. /kaɾɾo/ → káɾ.ɾo → káɾ.ro → [ká.ro]. On the other hand, strengthening in the syllable rhyme (21)b is an optional rule of emphatic speech, which Harris assigns to the postlexical phonology, ordered to apply after resyllabification. In addition to the rule of postconsonantal strengthening mentioned above, the analysis requires a separate rule of stem-initial /ɾ/ strengthening to account for the contexts in (21)c,d. For Hualde (1989, p. 824), stem-initial strengthening is an obligatory, early cyclic rule of the lexical phonology.

The crucial argument for complete resyllabification comes from the contexts in (21)e,f. Harris’s ordering of optional rhyme strengthening after postlexical resyllabification explains the prohibition against prevocalic [r] at the right edge of a word or prefix. At the point in the derivation when emphatic strengthening could apply, the taps in question are no longer in the syllable rhyme, having already resyllabified into the following onset. Because emphatic strengthening could still apply if such taps were to maintain their lexical affiliation to the preceding rhyme, resyllabification must be assumed to be complete.

Bradley’s (2020) monostratal OT account of the NCPS rhotic distribution turns out to be entirely compatible with an ambisyllabic representation of incomplete resyllabification. The analysis uses a perceptually-based markedness constraint on consonant duration contrast and two markedness constraints on the in-phase coupling of tap and trill gestures with a following vowel. Grounded in previous work on rhotics in Spanish (Bradley, 2005b, 2006b) and Catalan (Padgett, 2009) within the constraint-based Dispersion Theory of Contrast (Flemming, 2004), (22) rules out a surface contrast between the relatively shorter tap [ɾ] and the relatively longer trill [r] in any position that is not stem-medial intervocalic:

Bradley (2020, p. 247) initially relativized this constraint to the prosodic word domain. In keeping with the positional faithfulness constraint (13)a, the updated constraint (22) is relativized to the morphological stem domain. The definition is functionally motivated by Dmitrieva’s (2018) experimentally supported claim that differences in consonant duration are easiest to hear in the context of two surrounding vowels.

The analysis requires only two markedness constraints on in-phase coupling:

(23)a is violated whenever [ɾ] and a following vowel are coupled exclusively in-phase while belonging to the same morphological stem domain. ⁵ (23)b is more general, penalizing the in-phase coupling of [r] with a following vowel in any morphological context.

Tableau (24) gives an analysis of rhotic contrast and neutralization across six contexts:

Within each context, the input contains two highly idealized words or phrases, each tagged with its own subscript. These shapes stand in for actual words or phrases from the language’s lexicon but also represent phonological contexts in loanwords or neologisms. The analysis systematically considers three output candidates: (i) contrast between the tap and trill, (ii) neutralization to the tap, and (iii) neutralization to the trill. Combined subscripts indicate neutralization, or the merger of two words or phrases in the output. The ranking of Ident-(gesture) above the constraints on intergestural coupling predicts a tap/trill contrast in stem-medial intervocalic position (24)a, which corresponds to the existing minimal pair in (21)a, among others. Attempting a tap/trill contrast in any other context fatally violates perceptual spacing, as seen in (24)d,g,j,m,p. In each case, the two remaining neutralization candidates are tied on their faithfulness violations, which passes the decision down to the lower-ranking gestural coupling constraints. Since they refer specifically to in-phase coupling, *[ ɾ―V_Stem] and *r―V are inactive on rhotics that are coupled only anti-phase with a preceding vowel. The analysis predicts free variation between co-optimal (24)e,f, which corresponds to the non-contrastive variation observed in (21)b. Further gradient weakening can occur in the phonetic implementation of [ɾ] and [r] in this and other contexts (see Bradley, 2020, pp. 251‒253).

To understand the evaluation of candidates (24)g‒r, it helps to consider the morphological representations of examples (21)c‒f:

The ranking of *[ ɾ―V_Stem] above *r―V eliminates candidates (24)h,k, in which [ɾ] and the following vowel are coupled in-phase while belonging to the same morphological stem. Because their violations of lower-ranking *r―V are tolerated, candidates (24)i,l emerge as optimal, predicting neutralization to stem-initial [r] in (25)a,b. Although intervocalic word-/prefix-final [ɾ] is coupled in-phase with a following vowel, this vowel does not belong to the same morphological stem as the rhotic. Therefore, *[ ɾ―V_Stem] is inactive on syllable-initial [ɾ] in both word-final (24)m,n and prefix-final (24)p,q contexts. Because they satisfy the only remaining constraint in the hierarchy *r―V, candidates (24)n,q emerge as optimal, predicting neutralization to [ɾ] in (25)c,d.

Bradley (2020, pp. 246‒251) originally relativized *[ ɾ―V_Stem] to the morphological word domain. However, the constraint *[ ɾ―V_MWd] incorrectly excludes syllable-initial [ɾ] from the context in (25)d because the rhotic and the following vowel still belong to the same MWd domain, thereby overgenerating *supe[r]-ávido. Relativized instead to the morphological stem, *[ ɾ―V_Stem] becomes inactive in prefix-final prevocalic contexts, which correctly allows *r―V to favor syllable-initial [ɾ], as shown in (24)q.

Another piece of evidence that the morphological stem is the correct domain of *[ ɾ―V_Stem] comes from plural forms of rhotic-final singular nominals. An apparent restriction in the morphology of NCPS is that the nominal class of Ø-stems does not include roots that end in a trill /r/. Harris (1969, p. 51, 1983, p. 69, 2002, p. 96) interprets the lack of singular-plural alternations, e.g. flo[ɾ] ~ *flo[r]es ‘flower ~ flowers’ as evidence that a root-final trill is an underlying geminate tap /ɾɾ/. A final /e/ allows the geminate in /toɾɾe/ torre ‘tower’, as well as the root-final cluster in /kaɾne/ carne ‘meat’, to be properly syllabified. However, Baković (2009, p. 9) counters that positing a geminate interpretation is no better motivated than defining the relevant stems in terms of permissible final consonants. Other singleton consonants like /t/ must be excluded from this set, e.g. /bote/ bote ‘boat’ vs. */bot/, and a singleton trill /r/ would need to be excluded as well. An explanation for this morphological gap falls out naturally from our analysis, without the need to assume a geminate tap /ɾɾ/ in a language that otherwise lacks true geminate consonants. In line with (19)a, we postulate [[floɾ_Stem] es̺_MWd] as the morphological representation of the plural form flores, which allows us to invoke the same analysis that accounts for prefix-final prevocalic rhotics. Because the surrounding vowels do not belong to the same stem, the perceptual spacing constraint prevents a tap/trill contrast and forces neutralization. Because the /e/ of the plural suffix does not belong to the same stem as the rhotic, *[ ɾ―V_Stem] becomes irrelevant, thereby allowing *r―V to eliminate the trill in favor of the tap in the syllable onset, as in prefix-final prevocalic position (24)q.

Crucially, our analysis is able to ensure neutralization to morpheme-final prevocalic [ɾ] independently of how the rhotic is actually syllabified. Whether the rhotic is coupled ambisyllabically with both surrounding vowels (24)n or coupled only in-phase with the following vowel (24)q, it is the presence of a morphological stem boundary that crucially triggers a violation of Space-Dur/[ V_V_Stem] and also renders *[ ɾ―V_Stem] inactive. Alternative accounts of optional emphatic strengthening posit a rule (Harris, 1983, 2001, 2002) or markedness constraint (Colina, 2009, 2010; Roca, 2005) targeting individual rhotics in the syllable rhyme. Because it uses only in-phase coupling constraints *[ ɾ―V_Stem] and *r―V, our analysis does not have to worry about overgenerating a trill [r] when coupled anti-phase with a preceding vowel in ambisyllabic environments. Derived onset rhotics can be ambisyllabic (24)n, as long as gestural coupling constraints target only the in-phase coupling of a rhotic with a following vowel.

The main goal of this article has been to reconcile two opposing claims about the status of resyllabification in the Spanish phonology literature. On the one hand, a classic generative analysis of coronal fricative lenition in CS has long provided a compelling theoretical argument in favor of complete resyllabification (Section 2). On the other hand, recent experimental studies of NCPS have uncovered gradient phonetic differences in voicing and duration between derived and canonical onset /s̺/ (Section 3), which suggests that, at least for some speakers in laboratory speech, the structural difference between the two categories is incompletely neutralized. An explanatory phonological representation of the difference has remained elusive, until now.

Our approach to resolving these contradictory findings has been to start with insights from studies based on computational articulatory simulations (Section 4.1), which led us to propose a novel, competitively coupled split-gesture representation of ambisyllabic consonant shortening in AP (Section 4.2). We then showed how such a phonetically-based representation predicts the behavior of derived onset /s̺/ in NCPS and, with minimal changes in dynamical couplings, is further supported by Burroni’s (2022) TaDA simulations of the opposite phenomenon of ambisyllabic lengthening in Italo-Romance and Standard Italian (Section 4.3). Ambisyllabic gestural coupling then provided a simpler approach to coronal fricative weakening in CS and other varieties of contemporary Spanish, couched within monostratal, parallel OT, without ordered levels or strata, intermediate forms, or output-output faithfulness (Section 5).

As we demonstrated in Section 6, it is no longer necessary to assume complete resyllabification of prefix- and word-final prevocalic consonants in order to account for obligatory neutralization to [ɾ] in these contexts, pace Harris (1983, 2001, 2002). Henceforth, this pattern can be better understood as supporting our claim that in Spanish phonology, a perceptually-based markedness constraint restricts the rhotic duration contrast to stem-medial intervocalic position and that there is no rule or markedness constraint targeting individual taps or trills in the syllable rhyme. Both the aspiration of coda obstruents in CS and the emphatic strengthening of coda /ɾ/ in NCPS are compatible with incomplete resyllabification, represented as ambisyllabic gestural coupling, which in turn supports our definition of markedness constraints (14) and (23) on intergestural coupling.

The present study invites further laboratory and theoretical research. The mathematical implications of ambisyllabic competitive coupling for the shorter duration of the oral constriction of derived onsets in NCPS still need to be verified by explicit computational modeling in TaDA, following Burroni’s (2022) account of Italo-Romance. There is also a need for phonetic studies of resyllabification in varieties other than NCPS, using different consonants and experimental methodologies. ⁶ Finally, more work within the grammatical architecture employed here is necessary to explore the implications of ambisyllabic coupling for modeling other well-known segmental coda alternations of Spanish phonology, e.g. nasal velarization and liquid vocalization, and in a wider range of morphophonological contexts, i.e. prefixation and compounding.

Our analysis further supports a grammatical approach to the phonetics-phonology interface that does not require any transduction of distinctive features into articulatory gestures (Bradley, 2020; Smith, 2018; Walker & Proctor, 2019). Rather, the intergestural coupling graph of an utterance is phonologically optimized by OT constraint interaction and then phonetically implemented, where gestural overlap and reduction and universal aerodynamic principles further shape the speech output. In terms of the architecture of the grammar, phonology and phonetics are derivationally ordered components, but they nonetheless operate on the same representational primitives, namely intergestural coupling graphs.