Contiguity in prosodic words: Evidence from Spanish

2.1 The treatment of s aspiration as positional coda weakening

I base my analysis of the Spanish data on two assumptions. First, s aspiration is an instance of lenition, which is a general phenomenon reported crosslinguistically and involving a series of changes toward a simpler articulation of sounds in weak positions, along the so-called lenition trajectories (Lavoie 1996). These changes are phonetically induced (caused by the tendency of our articulators to reduce the degree of constriction and duration of sounds), and guided by such considerations as ease of production and perception. They can be both synchronic and diachronic, and lead to phonological changes (categoricity, alternations, influence on sound contrasts and on other processes). The phenomenon has been widely studied in literature (e.g. Kirchner 1998; see also Flemming 1995 for reflections on the perceptual basis of weakening). The second assumption is that the way segments behave across constituents is based on positional strength. Crucially, the syllable coda is a weak position, therefore we expect that the s will be more likely to weaken in the coda than in the onset (which is indeed the case). The concept of positional strength (Casali 1996) has been studied in conjunction with lenition e.g. by Honeybone (2008) and Ségéral and Scheer (2008) who distinguish four degrees of strength: word-initial onsets, word-internal onsets, word-final codas and word-internal codas. The latter two categories are especially relevant in studying Spanish s.

As noted by a reviewer, from the perceptual point of view, sibilant fricatives have robust acoustic cues and this is why they are alleged to resist changes, even in weak positions (Hayes et al. 2004; Henke et al. 2012). Nevertheless, s weakening has been reported in several languages, both synchronically and diachronically (e.g. in northern varieties of French, Corsican, Babine or Ecuadorian Spanish; Gurevich 2004). This involves either debuccalisation/loss or voicing, depending on the position of s. The process of debuccalisation itself is also well-attested. Typically, fricatives debuccalise to [h] while stops debuccalize to [?], the motivation being articulatory simplification or feature loss rather than exchanging one set of features for another (in both cases, the consonant simply loses place and retains the continuancy specification). Such changes have been observed in Navajo, Middle Chinese, English, Tauya, Páez and other languages. Furthermore, the glottal fricative [h] can be lost in phrase- and word-final position, as in Spanish, Nepali, Newari or Nez Perce. For an overview of lenition phenomena and their motivation, see the typological work by Gurevich (2004).

Consequently, despite the perceptual salience of s, we are dealing with positional weakening, which can be analysed from the point of view of sonority: high sonority segments are preferred in the coda (Sievers 1876/1893; Clements 1990; Baertsch 2002). Therefore, a sonority-based harmonic scale could be assumed by which the constraint banning s from the coda would be the second-highest ranked constraint after the one banning stops (e.g. Zec’s 1995 proposal). Indeed, such an approach is in line with the behaviour of Spanish obstruents, at least in the dialects discussed here. In Optimality Theoretic terms, Prince and Smolensky’s harmonic scales can be applied directly in the form of constraints corresponding to the (dis)preference of particular sounds in coda position (e.g. according to the proposed margin constraint hierarchy for codas, s is the second most dispreferred sound type and should be avoided).^[15] It must be pointed out, however, that the sonority approach to lenition has been criticized by some authors as it cannot explain certain consonantal lenition changes. The more adequate strength scales and lenition trajectories should be adopted instead, it is argued. See e.g. Honeybone (2008) and Broś (2015) for an overview of these issues.

Another perspective worth taking into account in the analysis of Spanish s is the Syllable Contact Hypothesis (Vennemann 1988), according to which rising sonority across syllable boundaries is avoided and the greater the sonority slope between the onset and the offset of the preceding consonant the better. This has been formalized differently by different scholars (see e.g. Gouskova 2004 or Baertsch 2002). Pons-Moll (2011) presents abundant evidence from Romance languages for the fact that margin slopes tend to be respected, which should be properly expressed by analytical tools used in phonology. Indeed, the lenition of s can be analysed as a response to the Syllable Contact Law in preconsonantal position. S aspiration and loss can be seen as an improvement of syllabic structure in terms of sonority. Nevertheless, prevocalic structures can be analytically problematic, and the approach is not able to account for prepausal behaviour of s. For this reason, I will not pursue this line of reasoning further.

To summarise this subsection, I assume that s changes observed in Granada and Chilean Spanish are instances of positionally-governed lenition and, in formal OT terms, repairs triggered by the coda condition against s – a specific instantiation of the general coda condition constraint. Hence, in Section 3, I use *S]CODA as the markedness constraint responsible for weakening.

2.2 A note on the underlying representations

The data presented in (1–4) show a variety of words with internal and final s. Not all of these words present alternations. Typically, nouns and adjectives ending in s and belonging to the group of irregular e-stems contain stem-final [s] alternating with [h] (in Granada) and [∅] (in Chilean). The example used in the data is the singular vez [beh] / [be] vs. the plural veces [be.seh] / [be.se] ‘time(s)’. Some other words of this type include: actriz ‘actress’, mes ‘month’, inglés ‘English’, capaz ‘capable’, etc. These alternations apply to both varieties under study. Additionally, alternations can also be observed in prefixes in Chilean (desconfiar vs. deshecho, as in (4)), but not in Granada. In all other cases, it is difficult to determine whether the s is underlying. Whereas in Chilean this would mean that both underlying /s/ and /h/ alternate with zero, which makes the analysis more problematic, in Granada it is possible to assume /h/ in the non-alternating cases. Nevertheless, I assume underlying /s/ for three reasons.

First, based on my fieldwork in Chilean and on my consultations with speakers of the Granada variety, I can assume that the /s/ is present in native speakers’ minds. There is variability in the pronunciation of s in careful speech, when the speaker is asked to read out a text slowly, repeat items or elicit them for a foreigner to understand (directions, etc.). From the point of view of learnability, native speakers must be exposed to the [s] variant when acquiring the language. This is probably dictated by orthography and scholarisation.

Second, I assume Richness of the Base (Prince and Smolensky 1993/2004), according to which both options are feasible and the grammar is responsible for selecting the correct surface forms regardless of the UR.

Third, I assume a free ride approach to morphophonemic learning, following McCarthy (2005b). According to this hypothesis, learners take alternations as the basis for their learning path and generalise the discovered patterns to nonalternating items, i.e. they derive all surface forms from the underlying representations derived based on alternations (unfaithful mappings). This approach is further legitimised by recent work on Catalan, another Romance language, which demonstrated independent evidence for the free ride learning algorithm (Pons-Moll and Lloret 2014; Lloret and Pons-Moll 2016). In my analysis, I use words that present alternations whenever possible (Chilean). Otherwise, I assume underlying /s/.

2.3 The treatment of prefixes and Spanish prosodic structure

In the analysis of examples (1–4), it might be argued that no opacity is to be expected in prefixed words as the prefix is syllabified together with the stem as a single prosodic word and hence deshecho behaves like semana ‘week’ or quise ‘I wanted’ where s occupies the canonical onset position and does not qualify for any weakening. The debuccalised word descalzar behaves analogically to the word esto ‘this’ where s is syllabified as a coda, driving lenition. Such a conclusion might be satisfactory for Chilean, but other dialects present different patterns (e.g. Granada).

Note that the resulting mismatch cannot be explained by solving the problematic status of the prefix. Indeed, Spanish prefixes do tend to behave differently than other affixes and have been considered to be separate prosodic words or ‘domains’ by some linguists (e.g. Wiltshire 1999,^[16] but also Face 2002, Shepherd 2003), but their accounts fail to present a uniform explanation for the diverse distributions of the fricative and its allophones across the Spanish varieties.

Another problem related to the treatment of Spanish prefixes lies in their place in the derivational hierarchy. Assuming a derivational (or stratal) approach, one must allocate them elsewhere than at the stem level. Whether they are attached at the same level as other affixes is another issue that shall not be further pursued here. What must be taken into account, however, is the similarity in the behaviour of prefixes and phonological words, at least in some dialects. Compare, for instance, the word deshecho [de.hé.tʃo] ‘undone’ with the sequence las ocho [la.hó.tʃb] ‘eight o’clock’ in the Granada dialect. Both present the same syllabification and debuccalisaton, suggesting a single driver for the process. At the same time, the prefix cannot resyllabify with the stem until the process of debuccalisation has applied, otherwise there is no motivation for it. Given that resyllabification must be post-lexical to generate debuccalised [h] in las ocho [la.hó.tʃo] in the dialect, the prefix must be assigned special status or deemed a different grammatical category. In a dialect which blocks debuccalisation across a prefix boundary before a vowel, on the other hand, the prefix behaves as any other affix in that it resyllabifies and leaves the fricative unaffected ([de.sé.tʃo] in Chilean). Here, the prefix seems not to enter the phrase level category, which makes it difficult to provide a uniform analysis of all the cases encountered across the Spanish dialects: why would the prefix have a different status for the purposes of one specific phonological process only? I try to resolve this issue in the next sections by resorting to the notion of contiguity and extending it to the supramorphemic level. Most importantly, however, I base my phonological analysis on the assumption that prosodic structure is the same for all Spanish dialects and strictly dependent on the morphological makeup of words. At the same time, prosodic mismatch (misalignment between the stem/morphological word and the prosodic word) obtains in phrase-level phonology, as well as in prefixed forms. Thus, the representational difference between the dialects discussed here lies in the resyllabification across a prefix boundary, which happens at the word level in Chilean, and at the phrase level in Granada, hence the similarity in the behaviour of prefixed words and word sequences in the latter. Also, since Granada admits optional deletion in some variants, postulating a different prosodic structure, e.g. that the prefix is a separate prosodic word, is not a viable option. With this assumption in mind, let us look at the morphological structure of Spanish words presented below.^[17]

1. Morphological structure building in Spanish^[18]

   individual words (sg)	vez ‘time’	[[vezRoot]∅]stem=MWd
   affixed words (inflection)	veces ‘times’	[[[[vezRoot]-e]Stem]-s]MWd
   affixed words (derivation)	carita ‘face’ diminutive	[[carRoot]-ita]Stem=MWd
   affixed words (prefixes)	descalzar ‘to unshoe’	[des-[[calzRoot]-ar]Stem]MWd
   	deshacer ‘undo’	[des-[[hacRoot]-er]Stem]MWd
   word sequences	esa cara ‘that face’	[[esRoot]-a]Stem=MWd
   		[[carRoot]-a]Stem =MWd

Against this background, the resultant prosodic structure may or may not match morphological concatenation. Following Bermúdez-Otero (2013) and others, I assume that certain affixes are root-based in Spanish, whereas others are stem-based. In Stratal OT this means that affixes can be attached at different derivational levels. Given the behaviour of prefixes in combination with aspiration, and with e-epenthesis (Peperkamp 1997), as well as the fact that they attach to independent words (Harris and Kaisse 1999) and have null impact on stress, I assume that they attach at the word level and are then subject to resyllabification against the underlying morphology. As a result, the word deshecho is syllabified de.she.cho, just as sequences of words are resyllabified together to form new prosodic words, e.g. es así ‘it’s like this’ syllabified e.sa.sí. At the same time, it must be remembered that in Stratal OT prosodic structure is erected at the first (stem) level and only then restructured in accordance with the grammar (the constraint hierarchy). As will be demonstrated in the remaining sections, differences in the constraint ranking account for the different behaviour of s across dialects. The prosodic structure of Spanish is presented in (6) below.

As illustrated in (6), prosodic structure does not necessarily coincide with the morphological makeup of the words in question (compare with the examples in (5)). For individual words in phrase phonology, I assume that the prosodic structure is assigned individually for each word (PW node), and then later on changed due to resyllabification (misaligned with the edge of the stem). This is exemplified in Optimality Theoretic terms in Section 3. As for the prefixed forms, the base is a stem analysed in stem-level phonology and hence parsed as a prosodic word. At the word level, when the prefix is added, another prosodic word is erected (PW’) that corresponds to the edges of the whole morphological word.^[19] Resyllabification and PW restructuring ensues either at the same level or at the phrase level, depending on the dialect. This is illustrated in section 3. Prosodic structure assignment is governed by alignment constraints: ALIGN-L(Morphological Word, Prosodic Word) and ALIGN-L(Stem, Prosodic Word). Provided that the Morphological Word is a higher-level constituent (as in example 5), the former constraint must be ranked above the latter to ensure the erection of the PW’ node.^[20]

1. Spanish prosodic structure (based on Chilean pronunciations)

   (a) stem level			(b) phrase level
   vez	‘time’	(beh)Pw	vez	‘time’	(be)Pw
   comí	‘ate’	(ko.mí)Pw	vez comí	‘time I ate’	(bé)Pw(ko.mí)Pw
   entré	‘entered’	(en.tɾé) Pw	vez entré	‘time I entered’	(bé)Pw(hen.tɾé)Pw
   calzar	‘shoe’	(kal.sáɾ)Pw	descalzar	‘unshoe’	(deh.(kal.sáɾ)Pw)Pw′
   hacer	‘do’	(a.sér)Pw	deshacer	‘undo’	(de.(sa.séɾ)Pw)Pw′

2.4 The role of contiguity

It must be noted that regardless of the variety, S does not admit uncompensated word-internal deletion.^[21] At the same time, deletion is quite common word-finally before a pause and across word boundaries (as in Chilean). Such a state of affairs seems to suggest that word-internal segments receive special protection as opposed to word edges, which can be explained by the notion of morpheme contiguity. Within the OT framework, this principle has been formulated as follows.

1. CONTIGUITY:

   Segments adjacent in the input should be adjacent in the output.

   (Kenstowicz 1994)^[22]

At this point, it is worth examining the exact definition of CONTIGUITY, which is quite problematic in OT. Initially, it was introduced as a constraint banning nonedge deletion and insertion in reduplicative languages. It was later extended to input-output correspondence relations, and although its definition does not mention any morphemic restriction, it has been traditionally treated as referring to morpheme contiguity. Although the need for such a constraint has been demonstrated in a variety of cases across different languages, the grammatical interpretation of its scope has not been taken up so far in the literature, with perhaps one exception (Lamontagne 1997). Thus, a theoretical debate over its domains of application would be desirable. This article aims at exploring this issue further with the analysis of the Spanish data. For the time being, it suffices to say that contiguity seems to play an important role in Spanish. On the one hand, it bans s elision in word-medial contexts, as in esto [eh.to] ‘this’ versus es que [e.ke] ‘that is’ in phrase phonology in Chilean. On the other hand, epenthetic processes confirm its relevance in initial contexts. The well-known process of e-epenthesis^[23] applies as a repair strategy to break illicit sC clusters, e.g. estrategia [es.tɾa.té. xja] ‘strategy’ or eslavo [es.lá.βo] ‘Slav’. Note that the epenthetic vowel helps create an additional syllable to which the illicit s attaches as a coda – an apparently nonoptimal strategy when confronted with inserting the e after the initial s to form a more optimal CV syllable as in *[se.tɾa.té.xja]. Given the fact that codas are strongly dispreferred in Spanish, the latter strategy would be more desirable were it not for the fact that it violates CONTIGUITY. This is further evidence for the higher status and strength of morpheme-internal positions as opposed to the weaker edges (which dates back to Trubetzkoy 1939; cf. Beckman 1999).

What is more, as illustrated by the data presented in Section 2, deletion is avoided across a prefix boundary in dialects permitting this strategy at word edges (desconfiado [deh.koɱ.fjá.ðo] ‘suspicious’ vs. es cómodo [e.kó.mo.ðo] ‘is comfortable’). This means that contiguity is respected also across morpheme boundaries, not only within morphemes. Given this observation, I suggest that the domain of contiguity should be extended to prosodic words, which is further discussed and illustrated in a tableau in Section 3.

3.1 Prefix treatment in Chilean under Stratal OT

After establishing the path of rendering the correct surface forms in the aspirating/deleting Chilean dialect, it is now time to determine how to deal with prefixed words. Note that in Chilean prefixed words behave like single morphemes in that no aspiration applies across a prefix boundary before a vowel. Thus, they cannot be analysed as sequences of words where aspiration applies. At the same time, the ban on coda s remains active in that prefix-final fricatives are aspirated before a consonant, but not deleted as in una vez comí. Therefore, the data confirm that prefixes do not behave like separate prosodic words or ‘special domains’ of any kind. Rather, in accordance with the proposed prosodic and morphological structure of Spanish prefixes (examples 5 and 6), they form part of a bigger constituent as bound morphemes. In prosodic terms, this is expressed in the form of a recursive prosodic word domain.

Given the Stratal OT analysis presented in the previous section, prefixes become problematic at the phrase level. The word level renders aspirated s ([h]), which cannot be protected from deletion. Thus, whereas a pre-vowel case of prefixation is neatly solved by syllabification at the word level, pre-consonantal cases escape this solution. Let us illustrate this in a tableau.

1. Word-level evaluation of deshecho ‘undone’

   /des + etʃo/	ONSET	CONTIG	*S]CODA	ALIGN-L	MAX(Seg)	IDENT(PL)	*h
   a.☞ (de.(sé.tʃo))				*
   b. (de.(hé.tʃo))				*		*!	*
   c. (des.(é.tʃo))	*!		*
   d. (deh.(é.tʃo))	*!					*	*
   e. (de.(é.tʃo))	*!				*

As demonstrated in (19), all cases of non-syllabification of the prefix with the stem result in a fatal violation of the highest-ranked constraint, while aspiration requires an additional, unnecessary violation of identity, thus the simply misaligned candidate (19a) is selected as optimal. Under such an analysis, however, descalzar ‘to unshoe’ would leave the word level with a debuccalised segment as the input to the phrase level: (deh.(kal.sáɾ)) – here, syllabification is not an obstacle as the onset position is already filled in the stem (calzár); s is obligatorily in the coda and has to debuccalise to satisfy *S]CODA. The desired phrase level candidate cannot surface under such an analysis: deletion is produced similarly to phrases such as una vez comí ‘once I ate’ although the placeless [h] should be retained. This is illustrated by the tableau in (20).

1. Unsuccessful phrase evaluation of descalzar ‘to unshoe’

   /deh + kal.saɾ/	CONTIG	*S]CODA	ALIGN-L	IDENT(PL)	*h	MAX(Seg)
   a.☹(deh.(kal.sáɾ))					*!
   b. (des.(kal.sár))		*!		*
   c. ☜(de.(kal.sáɾ))						*

Note that CONTIG does not protect descalzar in this case: it does not assign a violation mark to (20c) because of the morpheme boundary that intervenes between the prefix and the stem. As pointed out earlier, however, the exact definition of CONTIG is problematic in OT. Although it does not explicitly mention any morphemic restriction, it has been traditionally treated as referring to morphemes. Nonetheless, the Chilean case demonstrates that the grammar treats a prefixed word as a whole, barring deletion across a prefix boundary as if it were one morpheme. In this respect, prefixed words are treated in the same way as bare stems and suffixed words – they belong to the ‘word’ domain, which does not allow internal interference in the form of segment deletion. Elision of the underlying s is possible only at word edges (either before a pause or before a consonant-initial word). Inside words, aspiration is the last stage of s weakening. This is ensured precisely by CONTIG (cf. the word escondidos ‘hidden’ in example 9). To ensure that the same restriction is preserved in prefixed words, CONTIG must refer to the whole prosodic word.

1. CONTIG]PW: No deletion/insertion in the middle of a prosodic word.

   (adapted from Rubach 2011)

CONTIG]PW ensures that the input string is identical to the output structure in that no deletion or insertion takes place within it, protecting word-medial segments but not the edges. Such a constraint has already been proposed by Rubach (2011) to address insertion problems in Macedonian. Crucially, Rubach pointed to the problematic behaviour and status of clitics in this language, and to the need for a constraint that protects word-medial structures and mandates schwa insertion at the edge of the stem rather than inside it in both prefixed and nonprefixed words. Due to the fact that clitics do not form part of the PW in Macedonian, cliticised structures can exhibit internal insertion when analysed as a whole, but always at the prosodic word edge. This constitutes an important insight into the role prosodic domains higher than the syllable play in the analysis of phonological processes. The present analysis adds evidence from deletion to the evidence from insertion provided by Rubach.

Note that reference to prosodic structure is not problematic from the Stratal OT perspective as prosodic structure is erected at the first derivational level. The phrase level already contains information concerning prosodic word parsing at the level of the input. As argued in the previous sections, prosodic word structure is erected in the stem calzar ‘to shoe’, followed by the PW′ node after prefixation. With CONTIG]PW in place, the evaluation of descalzar at the phrase level no longer poses a problem.

1. Evaluation of descalzar ‘to unshoe’ with the use of CONTIG]PW

   /deh + kal.saɾ/	ONSET	SSG	CONTIG]PW	*S]CODA	ALIGN-L	IDENT (PL)	*h	MAX (Seg)
   a.☞(deh.(kal.sáɾ))							*
   b (de.(hkal.sáɾ))		*!			*		*
   c. (des.(kal.sáɾ))				*!		*
   d. (de.(skal.sáɾ))		*!			*	*
   e. (de.(kal.sáɾ))			*!					*

In (22), the correct output form is generated. (22e) is eliminated by CONTIG]PW. (22c) is defeated by *S]CODA. The other two candidates are suboptimal due to SSG, which requires that the sonority of segments rise toward the nucleus of the syllable.^[27] (22a) surfaces as optimal as it violates only the lower-ranked *h. Thus, the provided constraint ranking is able to account for all the Chilean examples, including prefixed words. With respect to the latter, note that treating prefixes as an integral part of the word in question (at least in prosodic terms) is the only viable option. No recourse can be made to morphology. One might suggest that prefixes are added at a different stage of phonology, alluding to their supposed special status in several languages, including English with its level 1 and level 2 affixes. Nevertheless, it results impossible to remove the obstacle in this case. Whether we add the prefix des-at the very beginning or at the end is irrelevant. The reason for such a state of affairs is simple: it is not the base (stem) that undergoes the change in question but the prefix itself; the stem remains unaffected. At the same time, the prefix is no solution in itself. The key to the analysis of Chilean aspiration and deletion effects is the prefixed word as a whole, apparently treated as one contiguous string resistant to deletion. This is effected by CONTIG]PW.^[28]

A reviewer points out that the prosodic structure of cliticised phrases as opposed to prefixed words should be discussed. Given the fact that Spanish prefixes have been argued to act like clitics, word sequences consisting of unstressed function words followed by host nouns may contradict the effect of the CONTIG]PW constraint. Note that such phrases as las normativas ‘the norms’ presented in (2a) above behave differently than prefixed words in Chilean. Namely, the underlying s is retained as [h] in prefixes (e.g. descalzar) but deleted in las normativas. In the latter case, the behaviour is typical of word sequences although the first word is a bound morpheme which cannot receive stress and cannot be assigned a prosodic word status. Nevertheless, the assumption about the activity of CONTIG]PW holds as the two types of bound structures differ in the way they are prosodified. I follow Selkirk’s (1995) position which dismisses intermediate constituents between the prosodic word and the phonological phrase (contra e.g. Nespor and Vogel 1986 who posit clitic structures). Selkirk assumes that function words and other bound morphemes present different prosodic structures depending on their phonological makeup, position and behaviour. The principal prosodic constituents are: syllable, foot, prosodic word, phonological phrase, and intonational phrase. They form a hierarchy and admit a special type of recursivity, e.g. in compounds and prefixed words where we have a PW′, as illustrated earlier. This stance is followed by e.g. Peperkamp (1996, 1997) and Lleó (2012). I assume that prefixes are affixed to the word and therefore attached to a recursive prosodic word node, whereas articles and similar proclitic structures, which display different behaviour (e.g. cannot form a foot together with the first syllable of their host, do not affect stress and are acquired differently by children), are attached directly to the Phonological Phrase. Thus, they do not form a PW with their hosts and the CONTIG]PW constraint does not apply to them. In Selkirk’s terms, they are free clitics (whereas prefixes are affixal clitics). To illustrate the difference, las normativas is prosodified (las(normativas)_PW)_PPH, as opposed to descalzar prosodified as ((des(calzar)_PW)_PW’)_PPH. The specific case of articles, such as las in las normativas is discussed by Lleó (2012) in terms of prosody, with arguments from L1 acquisition.

3.2 Prefixes in Granada-type dialects

After providing the correct evaluation of the Chilean data, at least one important task remains to be solved. The ability to account for other Spanish dialects will further legitimise the viability of the Stratal OT account presented in this paper. Let us then examine the Granada dialect data, where aspiration occurs both across a word and a prefix boundary before a vowel. The word deshecho ‘undone’ is put under evaluation in (23).

1. Word level evaluation of deshecho ‘undone’ with aspiration across a prefix boundary

   /des + etʃo/	ALIGN-L	ONSET	CONTIG	*S]CODA	MAX(Seg)	IDENT(PL)	*h
   a.☞ (deh.(é.tʃo))		*				*	*
   b. (de.(hé.tʃo))	*!					*	*
   c. (des.(é.tʃo))		*		*!
   d. (de.(sé.tʃo))	*!
   e. (de.(é.tʃo))		*			*!

In (23), candidate (23a) is selected as optimal. It presents aspiration in the coda. Note that the Granada dialect and the like require that resyllabification not take place at the word level. The prefix-final segment remains attached to the coda position of the first syllable. This is ensured by ALIGN-L, which is ranked above ONSET to mandate alignment. (23c) is eliminated by *S]CODA. A tie is produced in candidates (23a) and (23e) on ONSET. This is solved by MAX(Seg). Candidate (23a) surfaces as the winner and provides the input to the phrase level.

1. Phrase level evaluation of deshecho ‘undone’ with aspiration across a prefix boundary

   /deh.e.tʃo/	ONSET	CONTIG	*S]CODA	ALIGN-L	MAX (Seg)	IDENT(PL)	*h
   a.☞ (de.(hé.tʃo))				*			*
   b. (de.(sé.tʃo))				*		*!
   c. (des.(é.tʃo))	*!		*			*
   d. (deh.(é.tʃo))	*!						*
   e. (de.(á.tʃo))	*!				*

In (24), the desired candidate is chosen as optimal. Although (24a) ties with (24b) on ALIGN-L, (24b) is eliminated by the next constraint in the hierarchy – IDENT(PL). Candidates (24c–e) are eliminated by ONSET. Note that ALIGN-L is demoted below ONSET at the phrase level to ensure resyllabification. At the same time, MAX(Seg) stays in the same place, given that the Granada dialect does not allow deletion as a coda s weakening strategy. Consequently, the provided mechanism is able to account for all the presented cases, including the most problematic ones.

4.1 Kenstowicz (1996)

One output-based solution might be to propose a specific constraint that would apply to problematic cases. In his 1996 article, Kenstowicz introduces UNIFORM EXPONENCE, whose mandate is to “minimize the differences in the realization of a lexical item (morpheme, stem, affix, word)” (Kenstowicz 1996: 21). Thus, aspiration across a prefix boundary should be viewed as an instance of allomorphy minimisation: both deshecho and descalzar can be accounted for. As long as the UNIFORM EXPONENCE constraint is ranked above IDENT, the morpheme des-will always be realised in the same way. Nevertheless, Kenstowicz’s analysis collapses when he takes inflected forms into account (e.g. the plural meses ‘months’ realised as [me.seh] as opposed to the singular mes ‘month’ realised as [meh]). In such cases, the desired result is non-uniformity: Spanish presents a crucial discrepancy between the singular and the plural forms and no output correspondence obtains. Faced with these facts, Kenstowicz himself concludes that UNIFORM EXPONENCE must be restricted to the prefix des-, which definitely is a suboptimal solution in OT: morphemic liberty in this respect might lead to postulating an unlimited number of morpheme-specific constraints.

4.2 Colina (1997, 2002)

Colina (1997) makes an attempt at solving the same issue, providing a slightly modified analysis of the Granada data with the use of virtually the same set of constraints as Kenstowicz. She mentions UNIFORM EXPONENCE and another faithfulness constraint originally introduced by him: BASE IDENTITY. The second constraint requires that the components of complex output structures preserve their “original” forms. For instance, the word mes ‘month’ in mes entero ‘the whole month’ must have the same output structure as when produced in isolation, in this case: [meh]. According to Kenstowicz, such paradigmatic relations are necessary as they ensure maximal uniformity of inflected forms and morphologically complex words whose bases are independent words.^[30] Nevertheless, against this background, base identity cannot be ensured for the word mes at all times. For the plural form meses ‘months’, an unaspirated output form of the root is generated ([me.seh]), which results in a discrepancy between the different instantiations of the stem mes (mes, meses, mes entero). BASE IDENTITY must be mute.^[31]

The difference between the two constraints used by Kenstowicz (1996), and later Colina (1997), is that BASE IDENTITY compares outputs to independently occurring words (that form part of these outputs), while UNIFORM EXPONENCE compares pairs of bound and unbound allomorphs. In her analysis of Granada Spanish, Colina uses BASE IDENTITY when referring to inflected forms such as plural meses, and analyses prefixed words and phrases in terms of UNIFORM EXPONENCE. She explains that mes+es is subject to BASE IDENTITY because mes is an independent word. At the same time, UNIFORM EXPONENCE is mute in the evaluation of the word meses, as “we are not comparing one morpheme with its allomorphs as in [the prefix] des-; we are comparing competing analyses of the plural meses with its independently occurring constituent – the singular [meh]” (Colina 1997: 20). However, the rejection of UNIFORM EXPONENCE in plural forms lacks sufficient motivation. Apart from functioning as an independent word, mes is also a morpheme, the same that is encountered in the plural meses. Thus, unavoidably, we are analysing two allomorphs of the same morpheme (realized as [meh] in singular and as [mes] in plural). Whether this information is accessed or made use of by a given constraint is another issue. In this case, however, it follows from Kenstowicz’s definition of UNIFORM EXPONENCE that meses must be evaluated by this constraint (he refers to morphemes, stems, affixes and words). Second, it would be difficult to motivate the use of UNIFORM EXPONENCE with respect to bound morphemes (prefix des-) on the one hand, and isolated words in sequences such as mes entero ‘the whole month’ on the other, and exclude the plural meses at the same time. Furthermore, as Colina herself points out, a problem arises in dialects which aspirate s but not across a prefix boundary before a vowel, in which case the question is how to rank UNIFORM EXPONENCE to generate the correct result. Colina suggests that in such dialects the constraint should not apply to prefixes. This, in turn, would imply a language-specific restriction on a universal constraint. Thus, although Colina (1997) presents an interesting alternative to other OT accounts of Spanish aspiration, her analysis may result problematic in more complex cases such as the Granada dialect or varieties such as Chilean, in which deletion further obscures output-output relations.^[32]

Colina (2002) improves on both Kenstowicz (1996) and Colina (1997) in taking a prosodic perspective on output-output evaluation. She proposes a recursive PW structure for prefixed words and a corresponding output-output constraint mandating identity between prosodic words. This constraint applies both to Pws and to prefixes adjoined to Pws as a part of such recursive structures. As argued by Wiltshire (2006), this involves a rather complex procedure whereby both preconsonantal and prevocalic instantiations of the prefix have to be compared at once to determine whether the aspirated or non-aspirated variants are more optimal. As a result, the output structure of the prefix in prevocalic contexts is chosen based on a word in which the prefix is syllabified differently, i.e. the output [deh] in de. [h]hecho is selected based on analogy to [deh] in de[h].calzar. Although this is a possible line of reasoning in output-output phonology, it may be difficult to maintain vis à vis the Chilean data in which deletion is a legitimate process, but not in prefixed forms, and prefixed forms behave differently than sequences of free-standing words. In such a case, the PW structure might have to be changed to accommodate the data.^[33] The same applies to other Spanish dialects that present aspiration but do not pattern with the Granada variety. By contrast, the analysis presented in this paper involves uniformity in terms of prosodic structure building across dialects. Whether a rearrangement in this respect is legitimised by other processes or theoretically superior to the option presented here requires further study.

4.3 Summary of Output-Output accounts

In the light of the above, it may be concluded that base identity or surface form uniformity, however defined, does not constitute a valid solution for the Spanish cases discussed so far. Apparently, only prefixes, and possibly only the prefix des-, present the kind of uniformity argued for by Kenstowicz (1996) and Colina (1997). There is no need to resort to morphology by means of an affix-specific constraint. The assignment of a prosodic word or similar status to prefixes along the lines of Colina (2002, 2009) or Face (2002) is an improvement which makes it possible to account for dialectal variation in a more uniform fashion. Nevertheless, in view of the data presented in this paper, and the additional complication added by s elision, I argue that a solution proposing uniform prosodic representations across dialects and indirect morphological influence resulting from framework architecture (strata) is an insightful alternative to previous accounts. The ultimate solution to the problem presented by the deletion cum aspiration plus prefixes cases discussed here cannot be warranted by recourse to output-output correspondence only. By contrast, the analysis produced under Stratal OT is successful in that it requires crucially two stages of derivation – each with a separate ranking. Thus, prefixes are added at the word level and (in the Chilean case) deletion takes place at the phrase level, where syntax comes into place. Resyllabification is either there all along or activated at the phrase level, depending on the dialect (the latter case being true for the Granada variety). Naturally, dialects with no aspiration/deletion interplay constitute an even more straightforward case and do not challenge the analysis presented in this paper. What is more, the differences in the behaviour of prefixes and suffixes in Spanish are captured by the simple assumption that suffixation is a stem-based process that does not involve a recursive prosodic word structure.^[34]

A reviewer notes that a threefold approach might save the analysis of at least some of the presented data. Indeed, Granada Spanish can be addressed properly if a distinction is made between derivational and inflectional outputoutput relations along the lines of McCarthy (2005a). Given the fact that derivational relations are asymmetrical, as suggested by Benua (1997), the TCT model is the correct path to follow in affixed words and in phrase phonology as they both present cyclic effects and allow for clear base identification. For Spanish phrases such as unas enfermedades ‘some diseases’ or una vez entré ‘once I entered’ with overapplication of aspiration across a word boundary the base would be identified from the forms produced in isolation, with markedness governing aspiration at the cost of unfaithfulness to the input (unas [u.nah], vez [beh]). BASE IDENTITY could be used to determine the direction of the paradigmatic pressure: from the base in isolation to the word in a sequence. O-O faithfulness outranks I-O faithfulness. The second part of the analysis consists in reflecting the behaviour of inflected forms for which a base cannot be identified. Such paradigmatic relations are crucially symmetrical (McCarthy 2005a), which means that all the forms within a paradigm can influence each other. In this case, *s]CODA induces the change s => h in the coda (ve[h]), and the ranking IDENT(Pl) ≫ OP-IDENT(Pl) prevents it from overapplying in onset position in inflected forms (be[s]es). The OP-IDENT(Pl) constraint is an analytical tool used as a part of the Optimal Paradigms model and is used exclusively to govern inflectional relations. The resultant ranking for both prevocalic and preconsonantal forms in Granada Spanish is *S]CODA ≫ OO-IDENT(Pl) ≫ IDENT(P1) ≫ OP-IDENT(P1). As for the prefix, this approach requires yet another outputoutput relation: Kenstowicz’s UNIFORM EXPONENCE (or the related proposal put forward by Burzio (2005), METRICAL CONSISTENCY). Since the prefix cannot constitute a base, it has to be treated as an affix requiring prosodic and segmental uniformity across contexts. Thus, resemblance beyond the priority of the base problem has to be mandated by a separate constraint. This treatment of paradigmatic relations goes beyond any of the three output-output approaches presented herein (TCT, O-O with UE and BASE IDENTITY, and Optimal Paradigms). Rather, it is a combination thereof. It must be emphasized that such a model is able to account for the Granada data. Nevertheless, the Chilean portion of Spanish eludes this solution. First, its prefixes depart from the behaviour of word sequences, which can be potentially solved by a different ranking of the UE constraint. Second, Chilean admits deletion, which complicates the analysis of phrases. Assuming that deleted forms serve as bases in inflectional paradigms, a ranking paradox may ensue. Across words, the base cannot be reliably chosen. The only solution would be to allow for phrase-level paradigmatic relations at the phrase level only, which is equivalent to introducing strata. Otherwise, a sonority- or syllable contact-based model might be adopted in the form of constraints activated to protect word-final prevocalic s.

4.4 Prosodic analyses

Outside of the output-output realm, Wiltshire (2002, 2006) presents a prosodic analysis of aspiration, taking into account the behaviour of prefixes across the different Spanish varieties. She argues that prefix-final behaviour can be attributed to prosodic structure of the type presented in this paper (recursive Pw) and to the sensitivity of the s to prosodic boundaries. The crucial constraint responsible for s weakening is therefore Weak|Pw which mandates weakening before word boundaries. Naturally, in words such as (des(calzar)), the s stands before a Pw boundary and hence weakens to [h]. To account for word-internal aspiration in monomorphemes and suffixed words, Wiltshire (2006) proposes Weak|σ. The proposed solution neatly solves a number of problems in dialects presenting both opaque and transparent aspiration. One case, the Río Negro Argentinian Spanish, is similar to Chilean both in word-to-word and prefix behaviour, with the exception that it does not present deletion. Although Wiltshire notes that the constraint *h, when ranked above MAX(Seg), can account for deletion in those dialects that require it (Wiltshire 2002: fn5), she does not further explore the issue with the discrepancies in prefixes in mind. It appears, however, that when deletion is to be prevented in prefixed forms, this ranking is not particularly helpful and incorrect outputs are generated.^[35]

Torres-Tamarit (2012, 2014) presents an interesting alternative couched in the Harmonic Serialism framework (McCarthy 2008b, 2010). His analysis is based on morphology-prosody mismatches which obviate opaque interactions. Parse and alignment constraints govern stepwise prosodic structure building. They are also in interaction with the coda condition banning s at the right syllable margin and the constraint mandating place identity between the input and the output. In the course of the derivation, the relative ranking of alignment and parse constraints is responsible for the different orders of prosodic structure assignment and allows for prevocalic s debuccalisation in either transparent or opaque contexts. Torres-Tamarit illustrates this with examples of prefixed words and word sequences from different dialects. In the course of a harmonic serialist analysis, he is able to account for overapplication across a word and a prefix boundary. Thus, a compelling analysis is provided as an alternative to Stratal OT that requires no constraint reranking. Granada-type dialects do not pose a problem for this solution. Crucially, however, it does not include cases with deletion as a second weakening strategy. There is no way of generating the correct forms for Chilean. This is due to the fact that, as argued in the previous sections, MAX(Seg) must be ranked below IDENT(PL) to enable deletion instead of aspiration before a consonant or a pause. As we are dealing with phrase phonology here, the three words taken from the sequence una vez entré must be parsed into prosodic words separately and only then prosodified further into a phonological phrase to enable resyllabification. Yet with the low ranking of MAX(Seg) aspiration will always be suboptimal compared to segment deletion. Thus, aspiration cannot be generated. Conversely, if the words una vez entré are parsed together at an earlier step, then resyllabification is enabled and it will always be a better option to resyllabify the s than aspirate it at that point because it will allow for satisfying high-ranked ONSET without an additional violation of IDENT. Once the s is resyllabified, there is no motivation for it to aspirate at a later step (the ranking will not allow it). Thus, despite recourse to stepwise prosodic parsing which helps resolve opacity issues in various Spanish dialects, the modified version of Harmonic Serialism is unable to account for Chilean without constraint reranking. It remains to be seen whether another solution is found within the proposed mechanism.