As was illustrated in (2), with data from BLM2015, in NEC Catalan plural agreement can fail prenominally but not on the noun or postnominal modifiers. Other examples of prenominal missing s appear in (6), with a variety of prenominal elements that include articles, demonstratives, adjectives, possessives; the missing plural morph is realized in other dialects of Catalan (as [s] or [z] depending on the context). In the examples (6a–d) the noun is masculine, while in (6e) it is feminine.

But the plural morph does surface prenominally if it is preceded or followed by a vowel. Due to the fact that most feminine nominals end in a vowel while most masculine nominals end in a consonant, the vast majority of nominals without the plural morph are masculine. In the examples in (6) the plural morph fails to surface because the s would appear in the context C__C; that is, as the second consonant of a complex syllable coda.4 Some examples showing realized plural morphs in prenominal position after and before a vowel are given in (7) and (8), respectively.

As mentioned above, the phonological condition on plural morpheme realization (avoidance of a CsC cluster) holds only for prenominal position; in postnominal position the s surfaces regardless of the phonological context, as shown by the example in (2), repeated below as (9). Both the noun and the adjective bons ‘good’ surface with a final plural s in spite of the fact that it appears preceded and followed by a consonant in each case.

Another important aspect to bear in mind is that the s which does not surface in prenominal position must be the plural morph; any other final [s] or [z] does surface even if it is preceded or followed by a consonant, as the examples in (10) show, where the consonants in boldface are pronounced [s]; both the s in fals ‘false’ and the ç in dolç ‘sweet’ belong to the root.

Summarizing the facts, there are three conditions that must be met simultaneously for the non-realization of -s in NEC Catalan: (a) a phonological condition by which the missing s avoids a C__C context; (b) a syntactic condition by which the missing s is prenominal within the DP; and (c) a morphological condition by which the missing s is the plural morph.

In BLM2015, the interaction of phonological, syntactic and morphological factors in the failure of s realization is a consequence of the interaction between morphosyntactic constraints of the MAX and CONCORD family and other morphosyntactic and phonological constraints. The MAX constraints that are relevant are those in (11).

In BLM2015 MAX(SEGMENT) is violated in its standard sense, when a segment in the input is not present in the output, as is the case with plural s on the noun or postnominal modifiers, but it is also violated when a candidate, an output, has a morphosyntactic feature without the corresponding exponent, regardless of whether this morphosyntactic feature appears in the input or not. As for MAX(MPH), they consider it to be violated both when a morphosyntactic feature appears in the input, as in postnominal position, and also when a prenominal element surfaces without inflection. The different scenarios for MAX(SEGMENT) and MAX(MPH) violation and satisfaction are shown in (12) and (13), which correspond to BLM2015 (20a, b), with the exclusion of gender, for the sake of clarity. (12) illustrates an adjective in postnominal position, vell ‘old’, after having undergone concord with a plural noun in the syntax. The two MAX constraints apply as would standardly be assumed, with MAX(SEGMENT) being violated by the candidates in (12b, c), because the plural s of the input is missing; MAX(MPH) is violated only by (12c), because it has lost the feature [PLURAL].

It is in (13), with an indefinite determiner in the input (un ‘a’), that the additional meaning of the two MAX constraints comes into play. The input shows the root and, unattached, all the possible inflectional morphosyntactic features (here [SINGULAR] and [PLURAL]), with their corresponding exponents. This is a way to encode Vocabulary Items in Distributed Morphology terms. MAX(SEGMENT) is violated when a [PLURAL] feature in the output is not associated with s, as in (13b), but it is not violated when the output is a bare root, with no [PLURAL] feature, as in (13c); compare this case with (12a), where a bare root does violate MAX(SEGMENT) because the s morph of the input is not present in the output. MAX(MPH) is only violated by the candidate in (13c), a bare root, the idea being that any determiner should be inflected for number (and gender).

The tableau in (15) illustrates the relevant constraint interaction applied to the example aquest_ cabells llargs in (14) (part of BLM2015: (11a)).

The tableau in (15), in addition to the MAX constraints and CONCORD (relativized to number in NEC Catalan), includes other constraints. MATCH is a constraint banning contradictory morphosyntactic features. The constraint *PL (“no morphological expression of the plural”) belongs to the NOFEATS family (Samek-Lodovici 2002). Finally, *CsC is shorthand for the set of phonological constraints that ban an [s] in interconsonantal position (see BLM2015 for a justification of the constraint set and the constraint ranking).

The constraint MAX(SEGMENT) is violated twice by candidate (15d) for different reasons: the noun violates it because the input has a final s that is missing in the output; and the demonstrative violates it because the morphosyntactic feature [PLURAL] of the output does not appear with the corresponding exponent, s. The constraint MATCH is violated by the candidate in (15b), because the demonstrative is singular while the noun is plural. CONCORD(num) is violated by (15b) and (15c) because the demonstrative does not agree with the noun in plurality. The latter candidate, (15c), also violates MAX(MPH) because the demontrative does not have any inflectional feature. All violations of MATCH also constitute violations of CONCORD, while not all violations of CONCORD constitute violations of MATCH, given that a bare root does not agree with the noun, but it does not have contradictory features (compare (15b) and (15c)).

In (3) an example from dialects of Spanish was given that shows a feminine noun, arma [árma] ‘weapon’, triggering feminine agreement in postnominal modifiers but masculine agreement in prenominal ones. In what follows I first describe how this pattern arose and afterwards I summarize the account of it in BLM2015.

Historically in Spanish the feminine definite article la acquired the form el when a following common noun started with [á], avoiding a sequence of two identical low vowels, *[aá], a phenomenon that can be interpreted as a dissimilation caused by the Obligatory Contour Principle (OCP). In some approaches this form el has been identified with the masculine article, and hence as a change in morphosyntactic features, while in other approaches it has been analyzed as the feminine exponent undergoing phonological changes (see, for discussion, Plank 1984; Zwicky 1985; Harris 1987; and Kikuchi 2001, among others). I will assume, as reflected by the glosses, that the form el is the masculine exponent.

The example in (16a) shows the regular form of the feminine definite article, with a noun that starts with a consonant, casa ‘house’. The examples in (16b) contain different nouns that start with stressed [á]; only the definite article changes, postnominal modifiers surfacing clearly as feminine.

The change in the definite article does not apply when the following noun starts with an unstressed vowel, as exemplified in (17a).5 The avoidance of a phonological sequence containing two identical low vowels is not simply a phonological phenomenon. Nor does it apply when the definite article is followed by an element other than the noun, like an adjective or an adverb, (17b). Some nouns starting with [á] are exceptions and do not trigger the change, (17c).6 Moreover, the third person feminine singular accusative pronominal clitic la, which is identical to the feminine definite article, does not change to el when the following verb starts with [á], (17d).

This change of the definite article from feminine to masculine does not apply when the noun is plural, as shown in (18). This can be attributed to the fact that the presence of the plural morph s in the definite article breaks the [aá] sequence and hence avoids a violation of OCP.

Finally, in Standard Spanish, the change in the definite article applies only when the article is adjacent to the noun, not when a modifier intervenes, as illustrated in (19). In these examples the feminine article is not adjacent to [á]; the only [aá] sequence arises from the contact between the feminine adjective and the nouns starting with [á], but in Standard Spanish no change affects the adjective in this context.

The change to masculine that we have seen with the definite article is also found, in Standard Spanish, with the indefinite article (un), and quantifiers like algún. What is of interest here is an extension of this phenomenon, found mostly in colloquial varieties of Castilian Spanish, but banned by prescriptive grammars, by which all prenominal elements surface in the masculine while all postnominal modifiers are feminine. This phenomenon is found only with the closed set of about 25 common nouns starting with [á] like those that were exemplified in (16b). More detailed descriptions, additional data, and discussion of this phenomenon can be found in, among other sources, Nueva gramática de la lengua española (RAE 2009), Álvarez de Miranda (1993), Ambadiang (1999), and Eddington & Hualde (2008), who refer to the nouns that trigger the phenomenon as hermaphroditic nouns. Some examples appear in (20). In (20a), the masculine definite article is preceded by the universal quantifier todo ‘all’ in its masculine form, not the feminine toda; the phonological OCP context that would trigger the change is not present here. Something similar happens in (20b, c), where the definite article is followed by a prenominal modifier and there is no phonological trigger for the change. Of special interest is example (20d). In this case the noun agua is preceded by the invariable possessive su, without surfacing gender; nevertheless the preceding quantifier todo appears as masculine.

When the noun is plural, all modifiers, whether prenominal or postnominal, surface as feminine, as shown in (21a, b), the plural version of (20b, d), respectively. The original reason for the presence of full gender agreement in the plural was the lack of an [aá] sequence and was illustrated with the definite article in (18).

In the analysis of this variety of Spanish, BLM2015 assume that the triggering nouns are all feminine and therefore cause feminine agreement to postnominal modifiers in the syntax, as is the case with all other feminine nouns of the language. The fact that these triggering nouns all start with [á] is synchronically a coincidence, the phonological factor that caused the change not being relevant nowadays, as was illustrated in (20). BLM2015 attribute the lack of masculine agreement in prenominal elements to a rule of impoverishment, a Distributed Morphology operation (see Halle & Marantz 1993, among others); this rule, in (22), deletes the feature [FEMININE] postsyntactically only for this set of nouns.

The loss of the feature [FEMININE] from the relevant nouns prevents the assignment of this feature to prenominal elements, which will surface with default masculine. The derivation of a DP containing prenominal masculine and postnominal feminine is schematically shown in (23) (adapted from BLM2015: (55)).

In the tableau in (24), corresponding to an abstract example with a prenominal and postnominal modifier, like in (23), the constraint CONCORD is relativized to the features [FEMININE] and [PLURAL]. The additional morphosyntactic constraint *FEM (“no morphological expression of the feminine”) belongs to the same family as *PL. The input reflects the noun once it has lost its feature [FEMININE] through impoverishment (as in (23c)).

The cover constraint CONC (F,PL) is vacuously satisfied by all candidates because, on the one hand, the noun is singular (and the constraint is relevant for the feature [PLURAL]), and, on the other, it has lost the feature [FEMININE] through impoverishment before evaluation. The candidate in (24c) violates the constraint MAX(MPH) twice because the prenominal element is not inflected for either gender or number (it is a bare root). The constraint *FEM favors the surviving candidate with fewer instances of the feature [FEMININE], (24a), which surfaces with default masculine agreement.

As was mentioned earlier, when nouns like agua, arma or área, are plural feminine, agreement is found in all elements of the DP (see the examples in (21)). It is in this context that the constraint CONCORD(F,PL) becomes decisive, as shown by the tableau in (25).

Crucially, the impoverishment rule in (20) does not affect the set of nouns that include agua, arma, or área, in the context of [PLURAL]. Therefore, the presence of the features [PLURAL] and [FEMININE] on the noun, turns CONC(F,PL) into a fully relevant constraint, violated twice in (25c), where the feature [FEMININE] of the noun is not present on its modifiers. The ranking of CONCORD(F,PL) above *FEM causes the candidate with full agreement (25b) to beat the candidate with partial agreement (25a). The prenominal elements in (25a, c) also violate MATCH, not included in the tableau, because the values for gender between these elements and the noun are contradictory.

In Section 1 an example was given from Rasom (2008) from Fassan Ladin (Cazet subvariety, example (2)), in which a feminine plural noun triggers full feminine agreement in all the modifiers in the DP but plural agreement only in postnominal modifiers, not prenominal ones. The example in (2) is repeated below as (26).

The examples in (27) show the same adjective in postnominal and prenominal position; only postnominally does it agree in number with the noun.

Lazy concord, as Rasom (2008) calls the phenomenon, affects only the feminine plural, as in other Northern Italian varieties; with masculine nouns full agreement is systematically found.

Fassan Ladin differs from the Spanish dialect analyzed in Section 3.2 in two ways. Firstly, as was just shown, in Fassan Ladin lazy concord affects number when the noun is feminine while gender is the affected feature in Spanish, with masculine instead of feminine surfacing prenominally. Secondly, in Fassan Ladin all feminine nouns are affected, while in Spanish only a small subset of nouns trigger lazy concord. This means that no impoverishment operation is needed for Fassan Ladin. These two varieties differ from NEC Catalan in the sense that lazy concord is systematic prenominally; it does not depend on specific phonological contexts. Therefore no phonological constraint will be interspersed among morphosyntactic constraints. The constraints needed to account for partial agreement in Fassan Ladin have all been presented. The constraints CONCORD(F) and CONCORD(PL) which in Spanish have the same ranking and were hence presented as the cover constraint CONC(F,PL) must now occupy different positions in the hierarchy, because of the active role of the markedness constraint *PL (belonging to the same family as *FEM), which must be ranked above CONCORD(PL), CONCORD(F) never being violated. Two slightly different constraint rankings can derive the right surface outputs, given that in sentences like (26) the adjective pìcola can be interpreted either as a feminine singular form (pìcol-a_F-Ø_SG), which would violate the constraint MATCH, in addition to CONCORD(PL), or as a feminine form with no number value (pìcol-a_F), violating only CONCORD(PL). For reasons of space the tableau in (28) includes only candidates with the second interpretation and omits the constraint MATCH. The constraint CONCORD(F) has not been included either because all the candidates shown satisfy this constraint.

In (28c), there is a fatal violation of the highly ranked constraint MAX(SEG), because the [PLURAL] feature that was assigned syntactically to the postnominal adjective has been deleted.7 The candidate with full agreement, (28b), violates only one of the constraints, *PL, but it does so in more instances than the winning candidate, (28a). The constraint MAX(MPH) is violated by the prenominal adjective in (28a) and (28c), because it lacks a value for number; in (28c) there is a second violation of MAX(MPH) because the feature [PLURAL] that appears in the input for the postnominal adjective is missing in the output. The rankings CONCORD(F) >> *FEM but *PL >> CONCORD(PL) cause full feminine agreement but only partial plural agreement.

One observation I will not attempt to analyze here concerning Fassan Ladin (and also the Friulian variety discussed in sections 4.2 and 5, but not varieties like Gherdener, for instance) is that, according to Rasom (2008), with postnominal modifiers the noun itself can be plural or singular; when the noun is singular the meaning varies slightly. Adjectives in prenominal position, as in (29a) and postnominal adjectives only with the noun in the plural, as in (29b), receive a non-restrictive reading; the interpretation of (29a, b) is that all houses are small. Adjectives in postnominal position with the noun in the singular, as in (29c), receive a restrictive reading; in this case for example, the implication is that some but not all of the houses are small.

The data in (29a, b) follow essentially the same pattern as NEC Catalan or dialects of Spanish, with postnominal modifiers agreeing with the noun and prenominal ones lacking this agreement. (29c) deviates from this pattern: while the modifiers behave as expected (singular prenominally, plural postnominally), it is the noun that unexpectedly surfaces in the singular. The absence of number on the noun cannot be attributed to a postsyntactic impoverishment operation, because it is not a property of a closed class of items, like the class of nouns starting with [á] in Spanish. In Rasom’s (2008) analysis, the noun in (29a, b) has moved through NumP, where it gets the feature [PLURAL], to a relatively low position within the DP; for (29c) she suggests that the noun moves to a high position in the DP without passing through NumP, and hence surfaces in the singular. The lower versus higher position that the noun occupies in (29a, b) and (29c), respectively, would account for the differences in interpretation, but it would be difficult to justify why N would have skipped NumP in (29c) but not in (29a, b).

The examples from Fassan Ladin that were included in section 4.1 showed that when there is lack of plural agreement, in prenominal position, the modifiers (and the noun in (29c)) surface in the feminine singular form, a final -a. When there is plural agreement, the feminine plural modifiers and the noun surface as final -es. So in the plural there is not just the addition of an s but also a change in the quality of the preceding vowel. One variety of Friulian, Friuli Occidentale (henceforth FO), has a similar pattern of exponence, with the feminine singular ending in -a or -e and the feminine plural ending in -is, according to Rasom (2008) (a fairly similar pattern is described in Manzini & Savoia 2005 for the variety spoken in Montereale). An interesting difference with respect to Fassan Ladin is that in FO the output form in partial agreement is identical to the plural minus the final -s, that is, -i instead of the feminine singular -a or -e. Relevant examples are given in (30), where the exponents of inflection appear in boldface and the missing plural morph -s is signaled with underscores. (30a) illustrates the exponents of nominal inflection when the noun is feminine singular: final -a is found in the definite article, while the adjective and the noun have a final -e. In (30b) feminine plural full agreement (always a possibility in DPs) is expressed with final -is for all categories. (30c) illustrates the pattern of interest here: when the adjective precedes the noun, what surfaces prenominally in defective agreement is the plural form -is but without the final -s, not the feminine singular forms, shown in (30c′). Finally, when the adjective is postnominal, as in (30d), it shows full agreement.

Summarizing the facts, the asymmetry between prenominal and postnominal inflectional material present in NEC Catalan, varieties of Spanish, and Fassan Ladin, is also present in FO, but in this case, when defective agreement is found, it is not the feminine singular that surfaces, but the feminine plural form excluding the -s. According to Rasom (2008) this pattern is also found in early stages of the acquisition of Fassan Ladin, when children produce prenominal elements with final -e (corresponding to the plural form minus the final -s) instead of final -a for examples like (29a); that is, they pronounce pìcole instead of feminine singular pìcola or feminine plural pìcoles.

An additional fact that Rasom (2008) points out regarding FO is that, in addition to the position of modifiers within the DP, the presence of final -s seems to be phonologically conditioned: this consonant is realized prenominally when otherwise a sequence of vowels, VV, would surface. The -s thus has an anti-hiatus effect, as illustrated in (31), with examples from Rizzolatti (1988), cited by Rasom (2008: 169). In (31a) the final -s of the adjective surfaces because the following noun starts with a vowel; in (31b) the noun starts with a consonant and the final -s, which would surface in coda position, is missing. Hence FO resembles NEC Catalan in having lazy agreement depending partially on phonological requirements.

The phonological constraint that favors deletion of sibilants in (31b) vs. (31a), penalizes s in coda position. In a candidate like (31a), this constraint is violated by the noun but not the adjective (because the plural morph of the adjective is an onset). This constraint, which appears as *s] in the tableaux that follow, is not too different from *CsC for NEC Catalan.

Similarly to other Romance varieties, including Fassan Ladin, the sigmatic -s present in the plural can be assumed to be the exponent corresponding to the feature [PLURAL]. The preceding vowel, independently of its shape (a or e in the singular, i in the plural) can be considered a class marker in the sense of Harris (1991), a thematic element, as in Oltra-Massuet & Arregi (2005), or a stem formative, following Bermúdez-Otero (2006) (see also Lampitelli 2014 and references therein). In all these approaches the vowel is indirectly related to gender; it is a morph different from the plural. The fact that the quality of the vowel differs in the singular and plural is not uncommon in Romance, as Fassan Ladin, in Section 4.1, illustrates, with -a in the singular but -e in the plural (compare cèsa ‘house.FSG’ and cèses ‘house.FPL’). In Asturian, to give another example, there is an alternation between u and o in the masculine and between a and e in the feminine for most nouns and adjectives (compare llobu, llobos ‘wolf, wolves’ and casa, cases ‘house, houses’; Academia de la Llingua Asturiana 2001), the pattern corresponding to the feminine being the same as what we find in Fassan Ladin.8 We can assume then that final -a and -e in FO are allomorphs of the feature [FEMININE], and that the vowel -i that appears in the plural is also an allomorph of [FEMININE] which surfaces only in the context of [PLURAL]. The fact that in lazy concord environments the vowel -i appears means that morphosyntactically there is full concord (the constraint CONCORD is satisfied), with only the plural exponent missing.

Applying to FO the constraints as understood in BLM2015, however, gives rise to a ranking paradox, which means that the constraint set is not appropriate. This ranking paradox is shown in (32) by means of a combination tableau (see McCarthy 2008). In a combination tableau, violations for each constraint are shown, as in regular tableaux, but the constraints are not ranked. The intended winner appears in (32a). For all the losing candidates (losers), (32b–d), a W in a cell means that that constraint favors the winner; an L means that that constraint favors a loser. The right ranking is obtained if for any candidate the constraints can be ranked such that all L are preceded by at least one W. As the Ws and Ls in boldface highlight, the problem for BLM2015 is that, when considering (32b), the conclusion is that *s] must be ranked above MAX(MPH) (*s] >> MAX(MPH)), while, when considering (32d), the conclusion is exactly the opposite: MAX(MPH) should precede *s] (MAX(MPH) >> *s]). Adding to the tableau a constraint like *PL will not make any difference, because these candidates, like the winner, violate the constraint three times.

When ranking paradoxes arise, the solution is often to find an additional constraint that can be highly ranked and favors the winner. The solution proposed in the next section is based on a reinterpretation of one of the constraints in BLM2015.