Fricative–affricate alternations in Catalan

3.1 The classical rule-based generative analysis

The fact that, as shown in (4), all four palatal sibilants contrast in intervocalic position, but three of them neutralize to [t͡ʃ] in word-final position led Lleó (1970) to conclude that the segments occurring in intervocalic position correspond to the underlying forms and to posit a rule of final /ʒ/ affrication, a fortition-based analysis. The same analysis is assumed in later work, like Mascaró (1976), Wheeler (1979), or Bonet and Lloret (1998). The informal version of the rule that appears in (6) is the one given in Bonet and Lloret (1998: 107).

The rule in (6) precedes final devoicing, in (7).

If final devoicing preceded final /ʒ/ affrication, there would be no targets for affrication, because all /ʒ/ would become [ʃ] (e.g. /bɔ́ʒ/ → *[bɔ́ʃ]). ^[2] If a more general process of affrication were posited, followed by final devoicing, a wrong output would be derived for underlying /ʃ/, as shown in (8) with the examples in (4a), boig [bɔ́t͡ʃ] ‘crazy.m’, and (4b), fluix [flúʃ] ‘loose.m’.

Therefore, the conclusion is that the affrication rule must be specific to /ʒ/ and must precede the more general devoicing rule.

According to Mascaró (1986a), the domain of final /ʒ/ affrication is the word plus clitics, based on evidence from some dialects of Catalan. This domain corresponds to the clitic group in Nespor and Vogel (1986). In addition, the fact that, with enclitics, final affrication applies only when the enclitic starts with a consonant provides strong support for the idea that, within the clitic group, the target must be in coda position (compare fuig-hi [fú.ʒi] ‘run.away.2sg.imp=there’ and fuig-ne [fúd͡ʒ.nə] ‘run.away.2sg.imp=from.it’). We will ignore domain-related issues in the rest of this paper, but the insight that affrication takes place in coda position is a key factor in the analyses offered in Section 3.2. An important shortcoming of this view is that an analysis based on fortition makes it a typologically rare phenomenon. Fortition is usually found in prominent positions (like word-initial position or onset position) and not in weak positions (like word-finally or coda position). The same observation is noted by Wheeler (2005).

3.2 Trying to generalize final affrication in autosegmental phonology

3.3 Final affrication in optimality theory

In this section, we discuss final affrication within several models of OT, starting with a parallel version of the theory, and we reach the conclusion that the mapping of /ʒ/ into [t͡ʃ] can only be derived through the use of ad hoc constraints.

The fact that there is general final devoicing can be attributed to the ranking of the markedness constraint No-Voiced-Coda, in (15a), above the constraint that demands faithfulness to voicing, Ident(voice) in (15b). And the fact that in final position there is an affricate instead of a fricative is a violation of a faithfulness constraint against changes in manner of articulation, which we will represent here as Ident(cont).

While the ranking No-Voiced-Coda ≫ Ident(voice) accounts for final devoicing of all obstruents, some constraint should force the selection of an affricate over a fricative at the end of the word, a fortition constraint. ^[4] This fortition constraint cannot be general, because it should not affect alveolar sibilants, only palatals, since alveolar fricatives do not become affricates at the end of the word (recall examples like grisa [gɾízə] ~ gris [gɾís] ‘gray.f/m’, in (3b), where the alternation involves voicing, not manner). We will refer to this constraint as Fort(pal)##, which would be violated by non-affricate sibilant palatals in word-final position. The ranking Fort(pal)## ≫ Ident(cont) ensures that a word like boig [bɔ́t͡ʃ] (/bɔ́ʒ/) ‘crazy.m’ is realized with a final affricate, as shown in (16). In all the tableaux, a W indicates that the constraint being considered favors the winner (the optimal candidate), while an L indicates that for that constraint the losing candidate (loser) fares better than the optimal candidate (see McCarthy 2008).

However, the same ranking leads to overapplication of affrication of underlying voiceless palatal fricatives, as in fluix [flúʃ] (/flúʃ/) ‘loose.m’. This is shown in (17), where the intended winner loses.

The fact that, as argued for in Section 3.1, in rule-based approaches final affrication must precede final devoicing in a sort of counterbleeding relationship could lead to the conclusion that a serial version of OT, such as harmonic serialism (McCarthy 2010), might be more adequate to handle final affrication and devoicing, that is, through the derivation /ʒ/→ d͡ʒ → [t͡ʃ]. The derivation of boig is shown in (18) and (19). In the first step of the derivation, (18), the possible candidates include one with the voiced affricate, (18a), the faithful candidate, (18b), and a candidate with devoicing, (18c). A form with [t͡ʃ] is not a possible candidate at this point because in harmonic serialism at each step Gen allows only one change with respect to the input for each candidate, and [t͡ʃ] differs from /ʒ/ in voicing and manner of articulation. The ranking of the relevant markedness constraints must be Fort(pal)## ≫ No-Voiced-Coda, with Fort(pal)## also being ranked above Ident(cont).

In (19), step 2, the input is the winner of step 1, /bɔ́d͡ʒ/. A candidate with [ʃ] is not possible because it differs from the input in voicing and manner of articulation. At step 3, convergence will be reached, because it is not possible to obtain a more harmonic form; so the winner of the derivation will be [bɔ́t͡ʃ].

But due to the needed high ranking of Fort(pal)##, harmonic serialism also predicts affrication with underlying /ʃ/, as was the case with the parallel version of OT, illustrated in (17) with the word fluix [flúʃ] ‘loose.m’, for which a surface form *[flút͡ʃ] is predicted. A more specific constraint is needed to exclude final affrication in underlyingly voiceless palatal fricatives; these segments must be faithful to their underlying representation; they cannot change. Let us call this constraint Faith(ʃ) provisionally, to see its effects. ^[5]Faith(ʃ) must be ranked above Fort(pal)##. This constraint will be vacuously satisfied for the example in (18) and (19), but it is crucial for the derivation of final [ʃ], which must remain faithful to its input. This is illustrated in (20). Convergence is reached at step 1.

As a matter of fact, once the need for a highly specific constraint like Faith(ʃ) is acknowledged, the motivation for the intuitively adequate serial derivation is lost. The inclusion of the same constraint derives equally well all the forms in a parallel account. The tableau in (21) corresponds to (17) with the addition of this new constraint (for (16), the derivation of boig in a parallel model, Faith(ʃ) is satisfied vacuously because the input does not contain a /ʃ/). ^[6]

The set of relevant constraints can be simplified if the comparative markedness model of McCarthy (2003) is assumed, a possibility also discussed for final affrication in Torres-Tamarit (2016). Within this model, for any markedness constraint there are two variants: either the marked structure is not present in the faithful candidate, identical to the input (new violation), or the marked structure is also present in the faithful candidate (old violation). The relevant alternative constraint under this view would be _N*ʃ, in (22), where the subscript N stands for ‘new’.

In comparative markedness, the constraint _N*ʃ would have a counterpart _O*ʃ (where the subscript O stands for ‘old’), a constraint that would penalize [ʃ] when present in the fully faithful candidate. With _N*ʃ, the constraint Fort(pal)## becomes unnecessary and the desired results are obtained. The tableaux corresponding to boig and fluix appear in (23) and (24), respectively, with this new constraint set. In (23c), the candidate [bɔ́ʃ] contains a [ʃ] that is not present in the fully faithful candidate [bɔ́ʒ] and hence constitutes a new violation of *ʃ. This situation contrasts with the candidate (24a) [flúʃ], the faithful candidate. In this case _N*ʃ is not violated, even though the candidate contains a [ʃ], because this consonant is also present in the faithful candidate itself; it is therefore an old violation, not a new one. The affricate [t͡ʃ] for input /ʒ/ in (23) is the only possible output, because the underlying voiced consonant is disallowed in final position and the most obvious repair, a change in voicing, is disallowed by the constraint _N*ʃ. There is no need for a constraint explicitly forcing affrication. A more drastic change, the deletion of /ʒ/, would be ruled out by Max(segment), a highly ranked constraint not included in the tableaux.

This constraint ranking will yield the right results for all other fricatives and affricates. The tableau in (25) corresponds to an example with an underlying voiced alveolar fricative, gris /gɾíz/ ‘gray.m’. The winning candidate, [gɾís], satisfies the markedness constraint No-Voiced-Coda and the candidate with final affrication, (25c), is ruled out because it unnecessarily violates Ident(cont).

One concern is that the constraint _N*ʃ is highly specific, and the same can be said for the handy constraint Faith(ʃ) in (20) and (21). It is not an easy task to formalize either constraint with features and exclude from the formalization the corresponding affricate. We assume, following the tradition started by Sagey (1986), also assumed in Mascaró (1984) and Lloret (1992), that the difference between fricatives and affricates is that the former have the feature [+continuant], while the latter have the feature [–continuant] in addition to [+continuant]. Hence, it is difficult to translate into the relevant features a constraint like *ʃ; in addition to the place of articulation features and [–voice], the feature [+continuant] should be mentioned to the exclusion of [–continuant].

So far, we have considered the [ʒ] ~ [t͡ʃ] alternation as the result of affrication (and devoicing) on a final /ʒ/, and have shown that, in either rule-based models or models like OT, it requires highly specific rules or constraints. The attempts to view affrication as part of a more generalized fortition syndrome result in overapplication or underapplication, leaving aside the fact that fortition in weak positions (like word-final position) are quite rare typologically. ^[7]

4.1 On the development of [d͡ʒ]

Up to now, we have explored the possibility that we are dealing with a fortition process involving affrication of word-final /ʒ/. Though this approach holds from a synchronic view, from a diachronic perspective the process took place the other way around, i.e. [d͡ʒ] underwent deaffrication intervocalically ([ʒ], as in rŭbĕa>ro[d͡ʒ]a>ro[ʒ]a ‘red.f’) and devoicing word-finally (rŭbĕu> ro[t͡ʃ] ‘red.m’) (e.g. Badia 1981: 194, 213, 216; Veny and Massanell 2015: 275–276; Caro Reina 2016: 148–149; Recasens 2017: 210). ^[8] In fact, in conservative dialects such as most Valencian varieties, though devoicing also applies word-finally (ro[t͡ʃ]), the affricate [d͡ʒ] systematically appears intervocalically (ro[d͡ʒ]a), for which a single affricate phoneme /d͡ʒ/ and absence of /ʒ/ is the most likely interpretation (Jiménez 1999: 220; Wheeler 2005: 11). Hence, intervocalic deaffrication is a relatively recent innovation in the development of Catalan. The fact that in all dialects the affricate was retained in certain words (as in me[d͡ʒ]e ‘doctor.m’) derives from their different etymological origin. In Old Catalan, the reflexes of Vulgar Latin -by-, -dy-, -gy-, and -j- underwent deaffrication (as seen in ro[ʒ]a), while the reflexes of Vulgar Latin -d’c- and -t’c- were maintained as affricates (as in med’cu> me[d͡ʒ]e). This distinction was systematically marked in the spelling of Old and Middle Catalan, with the former represented as <g, j> and the latter as <tg, tj>. As several scholars have pointed out, this implies that the two reflexes were pronounced differently. Some authors conclude that <g, j> involved the fricative [ʒ] pronunciation, while <tg, tj> involved the affricate [d͡ʒ] pronunciation (e.g. Miralles 1984: 135–136; Farreny 1986: 126–127; Rabella 1998: 167–169). Other scholars go further and suppose that the different spellings were indicative of a single affricate [d͡ʒ] pronunciation in the former case and a geminate affricate [d͡d͡ʒ] pronunciation in the latter, with deaffrication applying later to single affricates only (Lamuela 1984, 1992; Caro Reina 2016: 266–267). According to Caro Reina (2016: 278–279), evidence for the geminated pronunciation comes from word line division in fourteenth century texts (word division is indicated by blank spaces in the following examples):

Word division involving the affricates <tg> and <tz> suggests that the affricates are realized as geminates as in met ge (metge [médd͡ʒə] ‘doctor’) and met zinat (metzin-at [mədd͡zinát] ‘poison.ptcp[m]’), respectively [as happens with other geminates (e.g. marabal lant, meravellant ‘marvelous’)]. Otherwise, word division would have resulted in *me tge and *me tzinat.

As happens with single consonants (e.g. ma rit, marit ‘husband) and falling sonority transitions (e.g. res ponie, responie ‘(s/he) answered’).

From a typological point of view, patterns of deaffrication are also well attested in other languages, which are typically interpreted as spirantization processes, that is, as weakening effects in lenition-favoring contexts (Kirchner 1998). For example, Florentine Tuscan Italian shows spirantization of voiceless stops and /g/ in intervocalic position word-internally and across words, with an optional intervening liquid or glide (26a); in the same context, affricates become fricatives (26b) (Giannelli and Savoia 1978; Schafer 1995: 65; Lin 2011: 374; Kirchner 1998: 253–256). Along the same lines, in Kabardian ejective affricates spirantize and in Chitwan Tharu /d͡z/ becomes [z] in intervocalic position (Kirchner 1998: 304, 307).

Lin (2011: 373–375) also discusses a CVC prefixing reduplication process in Nisgha (a Tsimshianic language spoken in northwestern British Columbia) in which stops and fricatives in the root are fully copied (27a). When the second consonant is a velar or uvular stop, or a coronal or lateral affricate, it becomes a fricative (27b). According to Shaw (1987) and Lombardi (1990: 419–420), this change also implies a spirantization process, which does not apply to labials because there is no labial fricative in the phonemic inventory of the language and hence the process is blocked by structure preservation (Kiparsky 1985).

In the following sections, we discuss two different interpretations of the present-day Catalan patterns from the alternative view of deaffrication as lenition, within the tenets of OT. In Section 4.2, we examine an analysis that follows the assumption that there are underlying single and geminate affricates (/d͡ʒ/ and /d͡d͡ʒ/). In Section 4.3, we present our analysis, based on the existence of a single affricate /d͡ʒ/.

4.2 Starting from single and geminate affricates

Wheeler (2005: 15–22) discusses three possible interpretations for the relation between palatal fricatives and affricates: interpretation A consists of deriving all occurrences from a single affricate input; interpretation B consists of deriving them from a single fricative input; and interpretation C consists of deriving them from both affricate and fricative inputs with the addition of a segment unspecified for continuancy depending on the environment.

In Section 3.3, we presented the problems associated with deriving affricate occurrences from a single fricative, a possibility that Wheeler also dismisses. As for the other two interpretations, he concludes that interpretation C seems more plausible for the voiceless palatal affricates, with the coexistence of input contrasts involving /ʃ/ (e.g. fluix /flúʃ/: [flúʃ] ‘loose.m’, fluixa /flúʃə/: [flúʃə] ‘loose.f’) and /t͡ʃ/ (e.g. gavatx /gəbát͡ʃ/: [gəβát͡ʃ] ‘French.m (slang)’, gavatxa /gəbát͡ʃə/: [gəβát͡ʃə] ‘French.f (slang)’), as well as neutralized contexts in non-postvocalic positions (e.g. xinès [ʃinɛ́s] ‘Chinese.m’, panxa [páɲʃə] ‘belly’). He also assumes that there should be further inputs with biphonemic sequences of /tʃ/ to justify cases of word-final vowel epenthesis in words like cotxe /kótʃ/: [kót͡ʃə] ‘car’ (vs. gavatx [gəβát͡ʃ]) due to the sonority increase towards the syllable edge between the two coda consonants (fricatives being more sonorant than stops); we will return to the issue of epenthesis in Section 4.3. Contrariwise, he considers interpretation A to be more plausible for the voiced counterparts, i.e. /d͡ʒ/ inputs and absence of /ʒ/ inputs. In his account, though, interpretation A for all voiced palatal fricative and affricate occurrences entails input contrasts between single affricates (/d͡ʒ/), leading to weakening not only in expected lenition-favoring contexts, such as intervocalically (e.g. boja /bɔ́d͡ʒə/: [bɔ́ʒə] ‘crazy.f’ vs. boig /bɔ́d͡ʒ/: [bɔ́t͡ʃ] ‘crazy.m’), but also in typically fortition-favoring contexts, such as word-initially and postconsonantally (e.g. gel /d͡ʒɛ́l/: [ʒɛ́l] ‘ice’, menja /ménʒə/: [méɲʒə] ‘(s/he) eats’), as the author himself admits (p. 16). Additionally, he is obliged to assume inputs with a geminate affricate /d͡d͡ʒ/, leading to a geminate realization (e.g. metgessa /méd͡d͡ʒɛ́sə/: [məd͡d͡ʒɛ́sə] ‘doctor.f’) and further word-final vowel epenthesis (e.g. metge /méd͡d͡ʒ/: [me͡d͡d͡ʒə] ‘doctor.m’), as well as allomorphy (/d͡ʒ, d͡d͡ʒ/) to justify the absence of word-final vowel epenthesis in words like lleig /ʎéd͡ʒ, ʎéd͡d͡ʒ/: lleig /ʎéd͡ʒ/: [ʎét͡ʃ] ‘ugly.m’ (vs. metge [me͡d͡d͡ʒə]), lletja /ʎéd͡d͡ʒə/: [ʎéd͡d͡ʒə] ‘lletja.f’. In (28) and (29), we sum up Wheeler’s preferred interpretations with representative examples in each context. ^[9]

Nevertheless, Wheeler concludes that, within the tenets of OT, given

the logic of Lexicon Optimization and ‘Richness of the Base’ […] all occurring allophones need an explicit account. Thus when [d͡ʒ] and [ʒ] alternate, the constraints need to be identified to favour the correct output whichever occurs in the input representation. And when [d͡ʒ] and [ʒ] are in contrast, the account ought not to depend on details of input representation, provided that the input contrast is ‘notated’ in some way that is reflected naturally in the output. (p. 22)

For that reason, he ends up interpreting surface affricates as biphonemic sequences of stop-plus-fricative (e.g. /dʒ/, /tʃ/) and derives affrication and gemination from sonority requirements (we will make explicit the sonority requirements at work in Section 4.3). Nonetheless, he does not provide a formal analysis for the alternating fricative–affricate cases like [bɔ́ʒə] ~ [bɔ́t͡ʃ]. He merely points out that although this kind of alternation has some aspects in common with stop spirantization, which he analyzes as an effect of Laziness in lenition-favoring contexts (“Lazy: Minimize articulatory effort”; Kirchner 1998: 38), it does not result from the same causes in the same conditions. He additionally emphasizes the idea that in cases with no categorical complementary distribution of two (or more) allophones (as is the case of fricative–affricate alternations), lexical representations are important (pp. 326–327).

Pursuing the idea of single and geminate affricate inputs, Torres-Tamarit (2016) suggests that while voiceless palatal affricates contrast in being fricative or affricate (/ʃ/ in, e.g. fluix, fluixa, and /t͡ʃ/ in, e.g. gavatx, gavatxa), voiced palatal affricates contrast in being single or geminate affricates (/d͡ʒ/ in, e.g. boig, boja, and /d͡d͡ʒ/ in, e.g. lletja). In (30) and (31), we sum up the interpretation he proposes for each context. Torres-Tamarit does not discuss the surface coexistence of words with final palatal affricates and words presumably provoking word-final vowel epenthesis in the same context, i.e. cases (28d) (gavatx vs. cotxe) and (29c) and (29d) (lleig vs. metge) in Wheeler’s analysis. All we can infer from his work is that it might be considered that words like cotxe and metge have an underlying word-final vowel and thus follow the patterns in (30b) and (31b) for gavatxa and lletja, respectively (i.e. /kót͡ʃə/, /méd͡d͡ʒə/).

Torres-Tamarit models a tentative analysis within the bidirectional phonetics and phonology (BiPhon) model (Boersma 1998, 2009), which distinguishes multiple levels of representations. For our purposes, it is sufficient to differentiate the underlying form (represented between ‘|…|’), the phonological surface form, and the phonetic form. BiPhon allows a strict division between the phonological level and the phonetic level, with different constraints operating in each mapping: faithfulness constraints relate the underlying form with the phonological surface form; markedness constraints evaluate the phonological surface forms; cue constraints relate the phonological surface form with the phonetic form from the auditory point of view while articulatory constraints evaluate the articulatory phonetic form; and sensorimotor constraints relate the auditory form with the articulatory form (Boersma 2009). In Torres-Tamarit’s analysis, /d͡ʒ/ is phonologically penalized by the markedness constraint */d͡ʒ/ (on the basis of being a complex segment), which rules out all phonological occurrences of /d͡ʒ/ in favor of /ʒ/ at the phonological level. Then, an articulatory phonetic constraint Strengthen evaluates the degree of affrication that best suits each position (based on data by Hualde et al. 2015; see footnote 8), with intervocalic position being the context that most severely penalizes fortition (i.e. affrication). The maintenance of the affricate in words like lletja is attributed to its geminate underlying input |d͡d͡ʒ|, which blocks deaffrication at the phonological level presumably due to the effect of geminate inalterability (e.g. |ʎéd͡d͡ʒə|: /ʎéd͡d͡ʒə/). Though he does not mention the case of lleig, one could presume either that in word-final position geminate affricates obligatorily simplify because of the impossibility of having geminates in codas (|ʎéd͡d͡ʒ|: /ʎéd͡d͡ʒ/: [ʎét͡ʃ], *[ʎét͡t͡ʃ]) or alternatively that there is allomorphy, in line with Wheeler’s proposal (|ʎéd͡d͡ʒə|: /ʎéd͡d͡ʒə/, |ʎéd͡ʒ|: /ʎéd͡ʒ/).

4.3 Starting from a single affricate /d͡ʒ/

So far, we have shown that all analyses face the problem of finding a matching solution for the voiceless and voiced sets ([bɔ́t͡ʃ] ~ [bɔ́ʒə], [ʎét͡ʃ] ~ [ʎéd͡d͡ʒə] vs. [flúʃ] ~ [flúʃə], [gəβát͡ʃ] ~ [gəβát͡ʃə]), as well as for cases that presumably entail word-final vowel epenthesis in the same contexts that others do not ([gəβát͡ʃ] vs. [kót͡ʃə]; [bɔ́t͡ʃ], [ʎét͡ʃ] vs. [méd͡d͡ʒə]).

Our approach starts from the assumption that geminate affricate realizations within words are best dealt with as derived, and not as a consequence of geminate inputs. As other scholars have noted (e.g. Wheeler 2005: 36–37), geminates are rather rare in Catalan, especially in the obstruent set. There are few examples with /dd/ and /bb/; the items mainly belong to learned vocabulary or are loanwords, which strongly tend to degeminate in the case of /dd/ (32a). There are no /gg/ cases (loanwords like nugget are pronounced [núɣet]). Nor are there geminate fricatives, since they have evolved into a degeminated realization (e.g. piscina [pisɩ́nə] ‘swimming pool’, adolescent [əðuləsén] ‘teenager’). Geminate sonorants are more common but also tend to degeminate, especially if they are not presumably derived from assimilation (32b). Contrariwise, as discussed in Section 3.2.1, there exists a gemination process that targets voiced stops /b/ and /g/ before /l/ in certain contexts (32c), but not affecting Valencian and some other western dialects. (In the examples presented in (32), the most common pronunciation is given first.)

Regarding the issue of word-final vowel epenthesis, different scholars have pointed out that, while there is more conclusive evidence for internal epenthesis (see, e.g. Lloret 2002; Wheeler 2005: 252–257; Pons 2011) and word-initial epenthesis (see Lloret and Pons-Moll 2016), there is no morphophonological evidence for true word-final epenthesis in Catalan (see, e.g. Dols and Wheeler 1996; Wheeler 2005: 251–252). The evidence for word-final epenthesis is inconclusive, mainly because there are words ending in the nominal class marker - <e>[ə] after well-formed codas (e.g. rude /ɾúd-ə/: [ɾúðə] ‘rude-m’, pare /páɾ-ə/: [páɾə] ‘father-m’, enorme /ənóɾm-ə/: [ənórmə] ‘huge-m’, which show class marker -/ə/ instead of the regular ‘zero’ class marker that masculine nominals typically have (e.g. fred /fɾɛ́d/: [fɾɛ́t] ‘cold.m’, car /káɾ/: [kár] ‘expensive.m’, ferm /fɛ́ɾm/: [fɛ́rm] ‘firm.m’). A more convincing argument for their non-epenthetic status, put forward by several scholars for Spanish (e.g. Harris 1987; Colina 2003; Bonet 2006) and that now we extend to Catalan, is that while word-initial epenthesis is systematic and applies to new borrowings (e.g. [e]striptease in Spanish, [ə]striptease in Catalan), word-final epenthesis is totally unproductive (e.g. single (applied to records and people) [sɩ́ŋgel] in Spanish, [sɩ́ŋgəl] in Catalan vs. native words like ingl[e] ‘groin’ in Spanish, cingl[ə] ‘cliff’ in Catalan). Since no evidence leads overwhelmingly to an epenthesis interpretation, we assume that all word-final vowels are lexical, and hence are present in the inputs.

The final observation to take into account is that the phonology of Catalan has nonetheless an affrication process (lexical and postlexical) that turns stop-plus-fricative sequences into affricates, in heteromorphemic contacts (e.g. gats [gát͡s] ‘cats’ vs. gat [gát] ‘cat’) as well as across words (e.g. barret xinès [bərɛ̀t͡ʃinɛ́s] ‘Chinese hat’ vs. barret [bərɛ́t], Chinese [ʃinɛ́s]). Our starting point, thus, is that the unmarked situation is to regard all instances of surface affricates as deriving from stop-plus-fricative sequences through sonority-related constraints, except for the fricative–affricate alternations (i.e. the [bɔ́t͡ʃ] ~ [bɔ́ʒə] cases), for which we propose a single affricate input /d͡ʒ/, which turns into [ʒ] through lenition intervocalically. ^[10] The examples in (33) and (34) illustrate the palatal sibilant distribution we assume for each context.

The analysis proceeds as follows. Three markedness constraints control sonority relations among segments: (a) the general sonority contour of the syllable, i.e. Sonority-Sequencing (35a); (b) the sonority profiles in syllable contacts, i.e. Syllable-Contact (35b), which penalizes rising sonority transitions across syllables; and (c) the sonority distances between the segments of a complex onset, i.e. Minimal-Sonority-Distance_Onset (MSD_Ons) (35c), which, according to Wheeler (2005), in the case of Catalan permits only sequences with a sonority distance higher than or equal to seven points based on the scale in (36). ^[11] The other relevant markedness constraints at play are No-Geminate (35d), which militates against geminate segments; Lazy_{V_V} (35e), which favors lenition of voiced stops and affricates in intervocalic position; and No-Voiced-Coda, already presented in (15a). Lazy_{V_V} targets single consonants but not geminates, since the extra C slot generated by gemination destroys the context for Laziness to apply.

The relevant faithfulness constraints at play are Ident(voice) and Ident(cont), already presented in (15b) and (15c), respectively; Max (37a), which militates against segment deletion; and Uniformity (37b), which penalizes coalescence, as in the case of a stop-plus-fricative sequence becoming an affricate.

The constraint ranking at work is presented in (38) and justified throughout the tableaux (39)–(46) (see the ranking lattice in (47)).

The contrast between boja (39) and lletja (40) is captured through different inputs: with an affricate in the former, /bɔ́d͡ʒ₁ə/, and a consonantal sequence in the latter, /ʎéd₁ʒ₂ə/ (henceforth, we add subscripts to the relevant consonants for the sake of representational clarity). Given an input with an affricate, /bɔ́d͡ʒ₁ə/ (39), Lazy_{V_V} penalizes the presence of a single affricate between vowels because the affricate is associated with a single slot, (39b); however, the presence of a geminate affricate, (39c), does not violate Lazy_{V_V} because the extra C slot of the geminate destroys the intervocalic context for Laziness. Nevertheless, candidate (39c) violates No-Geminate. The ranking of Lazy_{V_V} and No-Geminate above Ident(cont), which the grammatical candidate (39a) violates, is sufficient for the candidate with lenition to win. (Tableau (39) presents arguments for the ranking Lazy_{V_V}, No-Geminate ≫ Ident(cont).)

Given an input with a consonantal sequence, /ʎéd₁ʒ₂ə/ (40), Syllable-Contact rules out candidate (40b) because the syllabic transition presents an increase of sonority, while MSD_Ons rules out candidate (40c) because the stop and the fricative onset consonants [d₁ʒ₂] are too close in sonority (i.e. a distance of four points according to the scale provided in (36)). Candidate (40d), with coalescence of the two input consonants /d₁ʒ₂/ into an affricate [d͡ʒ₁₂], entails a violation of Uniformity and, crucially, a violation of Lazy_{V_V}. Max rules out candidate (40e), where the input stop is lost. Candidate (40f), with a geminate affricate [d͡.d͡ʒ₁₂], is ruled out by No-Geminate and violates Uniformity too. Hence, the winning candidate, (40a), though phonetically identical to the geminate affricate in (40f), maintains the stop consonant but affricates the fricative, [d₁.d͡ʒ₂], for which reason it only irrelevantly violates Ident(cont). ^[12] In our account, words like metge [méd₁.d͡ʒ₂ə] are analyzed like lletja, because, for the reasons mentioned above, they have a word-final lexical vowel too: /méd₁ʒ₂ə/. (Tableau (40) presents arguments for the ranking Syllable-Contact, MSD_Ons, Max ≫ Ident(cont).)

Note that the fused affricate segment in (40d) occupies one slot, while the correspondents in the input occupy two slots; therefore there is a violation of Uniformity. But the root nodes of the input and the candidate (40d) stand in perfect correspondence; hence there is no violation of Ident(cont). Contrariwise, in (40a), Ident(cont) is violated because the second segment has a [–continuant] feature in the output, which is not present in the input.

The same analysis applies to all other alveolar or palatal stop-plus-fricative sequences, whether they appear within words (homomorphemic or heteromorphemic) or across words, which in the case of the voiced sets end up with the same phonetic surface-geminate effect (e.g. pot jugar /pɔ́d₁ ʒ₂ugáɾ/: [pɔ̀d₁.d͡ʒ₂u.ɣá] ‘(s/he) can play’, horitzó /uɾid₁z₂ó/: [u.ɾid₁.d͡z₂ó] ‘horizon’, set zeros /sɛ́t₁ z₂ɛ́ɾuz/: [sɛd₁.d͡z₂ɛ́.ɾus] ‘seven zeros’).

The fact that in word-final position words with an affricate input, like boig in (41), end up having the same phonetic outcome as words with a consonantal sequence input, like lleig in (42), is due to the fact that candidates with a voiceless fricative, i.e. (41d) and (42f), violate Ident(cont). In (42), the candidate with a voiceless fricative and stop deletion, (42e), crucially violates Max. Candidates (42b) and (42d) violate Sonority-Sequencing. Additionally, in both tableaux, any candidate with a word-final voiced obstruent violates No-Voiced-Coda, as occurs with (41b), (41c), (42b), and (42c). (Tableaux (41) and (42) present an argument for the ranking No-Voiced-Coda ≫ Ident(voice); tableau (42) further demonstrates the ranking Sonority-Sequencing ≫ Uniformity, as does tableau (43) below as well.)

All other stop-plus-fricative sequences in word-final position become an affricate as an effect of Uniformity being ranked lower than Sonority-Sequencing and Max, as shown with the word gavatx in (43). The same holds for heteromorphemic sequences (e.g. gats [gát͡s] ‘cats’).

When the inputs display word-final single fricatives, as in fluix (44) and gris (45), there is no reason to turn them into affricates, because this change incurs a violation of Ident(cont), as (44b) and (45c) show. Candidates with word-final voiced obstruents, though they may be fully faithful like (45b), are ruled out by the high-ranked constraint No-Voiced-Coda. (Tableau (45) presents an argument for the ranking No-Voiced-Coda ≫ Ident(voice), as did the previous tableaux (41) and (42).)

Finally, stop-plus-fricative voiceless sequences in intervocalic position, such as the one in gavatxa (46), are realized with a single affricate (46a). In our analysis, Syllable-Contact dominating Uniformity crucially rules out candidate (46b) in favor of the winning candidate (46a); MSD_Ons dominating Uniformity favors (46a) when confronted with candidate (46c), with a complex stop-plus-fricative onset; Max dominating Uniformity gives preference to (46a) over a fricative output (46d); Ident(cont) dominating Uniformity favors (46a) over candidate (46e), which maintains the stop and affricates the fricative; and No-Geminate dominating Uniformity favors (46a) over candidate (46f), with a geminate affricate (by transitivity of domination, the ranking is Syllable-Contact, MSD_Ons, Max, No-Geminate ≫ Ident(cont) ≫ Uniformity). Words like cotxe [kó.t͡ʃ₁₂ə] ‘car’ are analyzed similarly, because, as noted above, they are considered to have a word-final lexical vowel: /kót₁ʃ₂ə/. The same analysis holds for voiceless alveolar sequences within words (e.g. lletsó /ʎət₁s₂ó/: [ʎət͡s₁₂ó] ‘thistle, dandelion’) as well as for similar voiceless sequences across words (e.g. barret xinès [bə.rɛ̀.t͡ʃ₁₂in.ɛ́s] ‘Chinese hat’, compra’t segells [kòm.pɾə.t͡s₁₂ə.ʒéʎʃ] ‘buy.2sg.Imp=yourself stamps’). ^[13]

The final ranking lattice is presented in (47).

To close, it is worth noting that in our account, the possibility of ending up having surface-geminate affricate outcomes derives from the presence of two-slot inputs (i.e. stop-plus-fricative sequences), while the possibility of having fricative lenition derives from one-slot inputs (i.e. the affricate /d͡ʒ/). Jiménez (1996, 1999) ends up positing a similar interpretation for Valencian varieties that present surface-geminate affricates for the alveolar series, which are derived from /ts/ and /dz/ consonant sequences (as in lletsó [ʎett͡só], horitzó [oɾidd͡zó]), but single affricates for the palatal series, which are derived from /t͡ʃ/ and /d͡ʒ/ (as in gavatxa [gaβát͡ʃa], boja [bɔ́d͡ʒa], lletja [ʎéd͡ʒa]).