Before showing this analysis, I feel it is important to clarify some things which might be troubling some readers. Specifically, that the cyclic nature of prefixes is arbitrary/ad-hoc/unmotivated.

Naturally, it would be handy if this cyclicity could be derived from a more general syntactic fact, that cuts across Spanish dialects, but I am not aware of one. However, I would like those readers in particular to consider the following paragraphs before repudiating this aspect of the analysis.

Firstly, this is formal description of the system, it is not supposed to be the “first cause” of the pattern. I do not suppose that the /s/ began leniting because of the chunking in (6a).

Secondly, the lack of independent motivation for the prefix cycle in (6a) does not make the analysis any more ad-hoc than positing different rules at specific strata (Harris & Kaisse 1999), or different constraint rankings at different prosodic levels (Broś 2018). The question there remains the same: why is this ranking/rule at this level? In fact, since in Broś (2018) the Chilean word-level is the grammar of the Granadan phrase-level, it shows that the explanation cannot be based in the inherent nature of the rules and levels. Just as for Harris & Kaisse (1999), there is no independent reason for delaying Resyllabification and Aspiration rules to the phrase level in Río Negro.

I would even go further and say that the chunking analysis in (6) is even less ad-hoc than these previous approaches because there is only one phonology applied to different chunks, rather than having both different chunks (different bracketing and prosodic levels) and a different phonology for each level.

In fact, I do not think that the inability to derive the chunking from the syntax makes the analysis ad-hoc. Convergent evidence, or deriving the chunking from the syntax are good for supporting the chunking hypothesis presented in (6), but lack of exterior evidence does not invalidate its own evidence available to the learner.

Again, this is not a problem because I am not assuming that the chunking is the “first cause” of the pattern. I assume that a gradient phonetic lenition became categorialised/phonologised as a formal phonological analysis. Naturally, the formal phonological analysis has to fall within the range of possible phonologies.

In this specific case it amounts to saying that the possibility of being cyclic or not is made available by Universal Grammar. Counterfactually, we could have been in a world where all affixes are cyclic (matching the boundaries between words) or none of them ever are (always matching word-internal conditions). However, it is demonstrable that prefixes can exist in either state.

Where this fact is not derivable by some other property, the learner establishes it through the data. This appears to be also how parameters work, since they are also linguistically variable properties that cannot be derived or determined only by UG and are set by the child’s encounter with the data (even if this is mediated by complex internal structure such as parameter hierarchies) (Roberts 2019). In this case, the formal choice would be for heads in the morpho-syntax to be: Cyclic <yes, no>.

If one wanted to continue to think of this analysis as merely a description and not an explanation, they are welcome, but this is not so much a criticism of the analysis, it is merely a difference in the use of the word “explain”, and it applies just as much to the previous analyses.

The presence and the phonetic interpretation of empty skeletal positions is regulated by some universal mechanisms. They were largely developed in the literature for handling featurally empty vowel-zero alternations such as those of Morroccan Arabic (Kaye 1990) and French (Charette 1991). Since then, these conditions have found their application in very many languages, see Scheer (2004) for some of these.

Empty skeletal slots are also subject to a deletion operation called Reduction. This is developed from (Gussmann & Kaye 1993); it removes an empty V-slot followed by an empty C-slot from the representation (see V2-C3 below). This results in a newly formed CV string (C2V3).

The application of Reduction is cross-linguistically parametric, see Ulfsbjorninn (2022), and crucially, it applies at the Vocabulary Insertion itself. It is not a surface level constraint. Immediately as the phonological string is formed, Reduction either applies or it does not, and the resultant string is fed to the phonology.

These strictly alternating C and V slots enter into two specific lateral relations: Government and Licensing. These relations formally define positional strength in phonology (Ségéral & Scheer 2001; Scheer 2004): Government (Gov) is a weakening force, phonetically inhibitory, while Licensing (Lic) is a strengthening force, associated with the phonetic enhancement of a position. Positions that are Gov’d are weaker for it, and positions that are Lic’d are stronger for it. Filled V-slots can issue both Gov and Lic (shown in (8)). Gov and Lic are issued from V-slots in a right-to-left manner. As shown in (10), Empty V-slots require Gov to remain phonetically uninterpreted (cf. the extension to vowel-zero alternations (Kaye 1990; Charette 1991)), this renders the post-coda position especially strong since this C-slot is Licensed and not Governed: [–Gov, +Lic]. We see this in (10a).

If, however, the V-slot preceding the C is filled (V2), it will not be governed, as such the C-slot (C3) will be weak since it is also Governed: [+Gov, +Lic]. This is shown in (10b).7

Empty V-slots, such as (V2) in (10c), cannot provide either lateral relation. The Coda position is not Governed, but neither is it licensed, and therefore it is weak: [–Gov, –Lic].

As discussed in (§3) above, the string/phase initial position is marked with an empty CV. Being an ordinary piece of skeletal structure (and not merely a diacritic), this initial CV also participates in the lateral relations. Since its V is empty, it will absorb the Gov of a V-slot, consequently rendering the initial C “onset” strong, we see this in (10d).8

A final distinction that can be made phonologically in terms of V-slots and their potential for lateral relations relates to their position in the string. There is an inherent representational difference between the final empty V-slots “nuclei” (FEN), of which there can only ever be one, and any of the medial empty V-slot “nuclei” (MEN).

The licensing potential of FEN with regard to MEN is variable across different linguistic varieties (Charette 1990; Cyran 2003; Scheer 2012; Faust & Ulfsbjorninn 2018; Balogné Bérces & Ulfsbjorninn 2023).

In Chilean we will see that (optionally) /s/ is completely lost [Ø] before FEN, unless this triggers a hiatus context (*VV) and in these cases the [s] is retained.

We will also see that the FEN in Buenos Aires Spanish is a licensor (just as it is in other varieties not specifically treated in this paper such as Llanito (Ulfsbjorninn 2022)).

The process of debuccalisation from a formal phonological perspective seems to be especially connected to the lateral force called Licensing. This is explicitly modelled for the n/ŋ-alternations we see in Galician (Ulfsbjorninn 2020). The way this representational, realisational process occurs is shown here.

The analysis is realisational in the sense that a phonological representation (surface form) is mapped to objects in the phonetic form for implementation. This spell out between phonology and phonetics makes reference to the phonological objects and their lateral relations.

Since there is a feed-forward spell out from phonological objects to phonetic implementation, the analysis here is completely consistent with a phonology-first, phonetics-second linguistic architecture familiar to Substance Free Phonology (Hale & Reiss 2008; Reiss 2017; Volenec & Reiss 2020, see also Scheer 2019; 2022; Chabot 2021).

The way debuccalisation works in this model, referencing lateral forces, is introduced using the case study: the realization of rhotics in Dominican Spanish (for an analysis of this in other Spanish and Portuguese varieties, see De Souza 2018).

The rhotic-realisation of Dominican Spanish follows the lateral forces discussed in (§4.3) very closely. We see a three-way, not a two-way, pattern of rhotic realization. The data comes from Jiménez Sabater (1975: 86–87) and Núñez Cedeño (1987; 1994: 31). The phonetic transcription follows from Willis & Bradley (2008).

A summary of the pattern is shown in the table beneath.

These environments correspond to the following lateral relations, which are converted into the phonology-phonetics interface conditions (spell out) for rhotics. The structures for these positions and the lateral forces operating on them are shown in (13) beneath. In these structures, the single underlying rhotic is shown with the symbol: ®.

A summary of lateral relations and their outcomes is shown in (14) beneath.

Following these conditions, a tap will be predictably banned from the initial position. This is because, being string initial, it will be preceded by the prefixed empty CV (15a). This inserted empty CV is a regular part of the phonological representation and so it will interact with the lateral relations, thereby converting an “initial r” into a [–Gov +Lic] C-slot (15b), thereby surfacing as a trill.

The specific way in which the lateral forces are being mapped to the phonetic implementation is shown beneath.

As will now be demonstrated, what is sketched for /r/ in Dominican Spanish (Author) is surprisingly similar to the Strict CV analysis of word-internal /s/ allophony in the s-debuccalising Spanish dialects; except that Chilean has the additional (optional) condition of deletion of /s/ in final position.

Now we see how word-internal s-debuccalisation operates formally. Recall the pattern for Granada and Chilean Spanish.

In Chilean, the FEN is associated with the complete non-interpretation of /s/ (a.k.a. deletion). Within words, in all the s-debuccalising dialects, we see the following distribution of lateral relations and their phonetic spell.

The pattern here shows that the phonetic spell out of the place feature of /s/ seems to be intimately connected to the lateral force of Licensing.

When the phonetic features of /s/ are being spelled out in preparation for phonetic implementation, the Place node of /s/ is only inserted if the phonological position is +LIC.10

The spell out conditions listed above are uniform for all s-debuccalising Spanish dialects. They apply non-problematically within words, and they are blocked systematically across words in all varieties, through the insertion of the empty CV discussed in (§3). The point of variation is the insertion of the empty CV at the prefix juncture, that is to say: some prefixes are cyclic and others are not.

This variation is summarised in the table beneath, positions that involve adding an Empty-CV are marked with +CV in the cell.

The same outcomes of /s/ are found in word-internal and prefix-junctures in Chilean, whereas in Granadan the prefix-junctures behave like word-junctures. The derivation of these environments is shown beneath.

The differences in the behaviour of the prefixes in Granadan and Chilean are assumed to follow from the presence or absence of a phase/domain-initial CV. This is a sequence of empty syllable structure that is added between prefix and stem in Granadan but not in Chilean.

As mentioned previously, in Chilean Spanish there is an extra process of deletion. This occurs to “pre-pausal” /s/ and to pre-consonantal /s/ at the word-juncture: otro[Ø]#puerto[Ø] “other ports”.

In this same context, but before vowels, /s/ is not deleted, instead, it is merely debuccalised to [h]: una#ve[h]#entré ‘once entered’.

The extra step of deletion occurs both in domain-final position and in the context of a pre-consonantal word-juncture. I unify these by equating being domain-final and pre-domain-initial. In these contexts, /s/ unlinks from its C.

Take the UR in (28a), the steps of unlinking and Reduction are simultaneous, I show them separately in (28b and c) for the reader’s convenience. Once these processes have occurred, the string is fed to the phonology. This gives the phonology a string with an unlinked /s/ and there is no phonological motivation to relink it, the computed form is shown in (28d). Unlinked (i.e. deleted) segments are shown greyshaded.

Since this is a feed-forward derivational approach to phonological computations, the /s/’s association line is lost an earlier step and since there is no phonological motivation to relink it, it stays unliked. The opacity here shows the difference between this and surface-oriented phonological approaches.

Crucially, however, when this same word-juncture context comes before a vowel-initial stem, it leads to a different outcome. Precisely because there is a phonological condition that encourages the non-un-linking or re-linking of the word-final /s/: the vocalic hiatus that would ensue from delinking /s/.

This computation results in a form with a linked /s/, despite its UR (C3) occurring in a pre-phase-initial position (CV4), that would usually entail the unlinking of /s/ (29a).

Unlinking, in this context, however, creates a VV hiatus (Charette 1991), shown greyshaded in (28b). To prevent this, /s/ is relinked to its position (C3) (29d). After the computation and the application of Reduction the /s/ in C3 is still found before an empty Gov’d V-slot (V4). Since this is effectively still a coda-position, the /s/ in C3 finds itself in a [–Gov –Lic] context. As such, it is obliged to surface without phonetic place features: [h]. This outcome is entirely analogous to the Granadan prevocalic prefix-juncture.

The dialects of Argentinian Spanish reported by Harris & Kaisse (1999) are similar to the Chilean presented in Broś (2018). In fact, the dialect of Río Negro appears to be identical to Chilean, except that it does not have the extra deletion rule (unlinking of /s/ in final position).12 The Buenos Aires Spanish, however, shows a much more important difference with Chilean (which it still otherwise resembles).

Buenos Aires possibly completes the positional s-debuccalisation typology since it appears that even its word-junctures are treated like word-internal positions.13

In Buenos Aires, for the purposes of s-debuccalisation, not even word-boundaries are distinguishable from the word-internal position. As such, I propose that there is no initial CV across any boundary in Buenos Aires Spanish. I am not aware of any direct evidence for this claim, however, since its typological pattern is so elegant, it would be interesting to find out whether there is in fact, convergent evidence for this hypothesis. At least, the claim is easily falsifiable.

The derivation of s-debuccalised items will therefore be the same as that of /s/ word-internally and at the prefix-juncture in Chilean (§5.2.2). Additionally, the word-juncture will receive the following derivation.

This means that across the word-juncture and before a vowel-initial stem, there will be Reduction (just as there is across the prefix-juncture). Since no initial CV was inserted at the juncture the /s/ will find itself after the computation in C2 before a filled V-slot (V3). Therefore, C2 will be [+Gov, +Lic]. Consequently, unlike in Granadan, Río Negro or Chilean word-juncture, the /s/ will surface as [s].